Thanks a lot. I have been trying to train a model for a week and finally I found this gem. I was having an issue regarding numpy: AttributeError: module 'numpy' has no attribute 'int'. The problem was with the latest version of numpy. I ran "pip install --upgrade numpy==1.23.5" and it fixed the issue.
I followed the steps that you mentioned in the video and I tried 4-5 times to train the model but in the end, after training it did not detect Jack Sparrow's face in the image when trying to give an input image
Hello, thanks for this video, straight to the point and no time wasted, I just wanna ask after the first training of custom dataset. how do i add more dataset unto the first trained custom dataset yolo? do i use the best weights if i wanna continue adding more dataset to improve accuracy and more classes? becuz i wanna train my yolo while it retains the previous data is was trained with without starting from scratch, so i was wondering if u could help me with that. Thank you alot :)
Thank you so much, I’m doing my final year project and this video helps me so much. But i have a question. If i trained it to detect 1 object after that i want to train it to detect another object but still keep the previous one. How can i do that?
Hello. I run it on laptop with RTX 3060, 6Gb vram. But it runs only if batch-size is 1. Otherwise it doesn't work, an error occurs: RuntimeError: CUDA out of memory. Tried to allocate 1.92 GiB (GPU 0; 6.00 GiB total capacity; 545.16 MiB already allocated; 3.80 GiB free; 588.00 MiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF. I use 2000 images for training.
Hi Sir, as told you in the recent past, thank you so much for your effort producing this awesome minicourses. I'd like so much if you could produce a Yolov7 or Yolov8 detect video for Raspberry Pi. What's your thought regarding this idea?
Great video...! I followed the same steps as in the video but ended with no detection - I have trained 10 images and labels. May I know what missing in the detection?
@@TheCodingBug oh, well the ram information is less relevant. What's the VRAM on the 1060? It would seem my 3060 desktop gpu is an ideal choice as I can train yolo on it in a relatively shorter time!
@@TheCodingBug after some reasearch it seems that the VRAM being 12GB is a real boost to the training. The fact that I can overclock it is also a bonus.
If you have a stero camera you can measure detected object distance. Yolo is giving coordinates of objects. Stero camera can use this coordinates and can take distance. We are working on autonoums car and we are detecting traffic sign and their distance for writing best algorithms.
Please make a video on how to custom train already custom trained model for example : I trained model for detecting trash and wants to improve the accuracy, so I again want to train the model with different dataset
Very clear! Such a nice video :) I just have a few questions... Would it be possible to do the inference with the pretrained weights (for instance, base yolov7) detecting only one class (for instance, person; or car)? In case it is, could yo please explain how? Regards
Yes it's possible. Follow pretrained yolo v7 tutorial (if you haven't already) and use flag --classes 0 to detect person class only. Detail about this flag is discussed here th-cam.com/video/K_OGcfXwskc/w-d-xo.html
Thank you so much! Now I am facing another problem... For my custom dataset, my images are non squared, they are 1280x720. I saw a flag "--rect" in order to show I am not using squared images for training. How should I define the "--img-size" flag? I don't think it is just typing 1280 720; but I don't find anything relevant surfing on the Internet.
Hi there @TheCodingBug! First of all thank you for this Video! I have two Questions: 1. Can you maybe make a deep dive Video for this? I mean i could perform the custom training. But it still just runs in cmd-window. Can you maybe make a detailed guide on the main-function and which parts do what? . For example which lines to modifiy in the code if i want to for example modifiy the output of yolo? In v3 i wrote a function that only outputs sth if a person is detected , but no text output for anything else. How to do this in yolov7? 2. Also im highly interested in transfer learning. In this video we made a custom model with just one class. But how can i for example add new custom detections to the already existing coco-model ?
Hi, I am having this error RuntimeError: CUDA error: no kernel image is available for execution on the device Whereas my cuda and pytorch versions are the same Name: torch Version: 1.11.0+cu113 I had 11.6 but i changed it to 11.3 now it should work but its not and iam stuck now. EDIT: I have gtx 770 which is old. But i want to run yolo can i run only on cpu ? As i have installed only cpu pytorch but it gave this error: AssertionError: CUDA unavailable, invalid device -1 requested Please Assist ASAP
Could you please explain the steps needed to train more on top of initial training? What steps to take after doing all the steps in this video(ways to move through yolov7-custom folder/anaconda prompt)? What files need to be deleted/altered? Generally, what do i need to do to train again to add more classes/img’s/annotations? Thank you very much.
Hi, this is a very nice video man! I have a question, is it possible to convert this model to TFLite following the same steps that you did on YOLO v4 video? Thanks a lot!
hello great video. Question: how the official YOLOv7 wants the bounding box format to be? Just like COCO format [xmin, ymin, width, height] or like previous YOLOs [xcenter, ycenter, width, height]?
Thank you for this amazing video. But, I am having some problems. When I run detect.py, no bounding box appears. Where is the problem coming from? I have Detectron and OpenCV installed. Are these necessary to visualize bounding boxes? I hope you can help me.
Detectron is not necessary. Opencv is. With regards to no bounding box, check precision and recall graphs in training folder (where best.pt file is located). If it has really small values, the model is struggling to learn and you need more data for training.
@@TheCodingBug Actually I wanted to test the detection with pre-trained weights by calling "python detect.py --weights yolov7.pt --source inference/images/horses.jpg --img 640". But, no bounding boxes appears in horses.jpg. I have approximately 2000 images for 4 classes. I trained the model in YOLOv5, thanks to your videos. I got 0.65-0.70 mAP. I trained the model in YOLOv7, but precision and recall stayed under 0.001 too low. I guess the problem comes from former issue. Is there any idea to solve this?
@@tugbayildizli3828 Check the other video for pretrained model. th-cam.com/video/n2mupnfIuFY/w-d-xo.html See if it works properly. If not, try updating Nvidia drivers for GPU usage or try CPU device instead of GPU.
@@TheCodingBug Hi SIr, I'm on the same situation. What does it means "...you need more data for traing"? Maybe more images (I have 57 images on Train folder). Thanks in advance
Hi there... can we have a conversation about object detection? would love to show you what we are building and to understand how OD can boost our project.
Hello brother , can you please let me know what is the software you used to modify the official files ? At 7.25 . Please reply soon , I need it urgently for my University Project
hello sir. thank you for the explanation. i try to deploy with my 30 minutes video duration but turns out the detection video becomes 16 minutes. does it related to fps?
Hi. I’m having an error while training. AssertionError: Label class 15 exceeds no=1 in data/custom_data.yaml. Possible class labels are 0-0. Not sure how to solve this.
Hello, I am trying to train a model on custom plant images the model gets trained but does not detect correctly and makes no bounding box, is it because of underfitting, because I have only 60 images of a plant specie, is there any way to train the model with low number of images
hello could you please help me, i train dataset but train also show 0 label in the box. and when i try to test, there is no bounding box label. whats wrong with my dataset ?
Hello good sir, I'm trying to train multiple objects at once but I'm getting "AssertionError: Label class 2 exceeds nc=2 in data/custom_data.yaml. Possible class labels are 0-1". I've already change number of classes to 2. What have went wrong here?
Hi, nice video. Really Helpful. But I have a problem when training. I have this error: MP: Error #15: Initializing libiomp5md.dll, but found libiomp5md.dll already initialized. I searched around and they say the solution is to install all packages without MKL support: conda install nomkl is this a good solution? I mean is it because my CPU is old? It is Intel(R) Core(TM) i7-7700 CPU
Just an update in case someone had the same issue as me. I found out what the problem was. It is because of torchaudio package which was installed on my env. I noticed that in this video you remove torchaudio from the installation whereas in one of your other first videos you just copied the line from pytorch website which has torchaudio in it. I basically remake the env without torchaudio and everything works fine. Thanks again for the well made tutorial.
at 3:03 I get the error " name 'labelImg' is not defined". I installed the labelImg fine but when I want to open it by writing "labelImg" its doesn't work.
I am getting this type of error what could be the reason assert mlc < nc, 'Label class %g exceeds nc=%g in %s. Possible class labels are 0-%g' % (mlc, nc, opt.data, nc - 1) AssertionError: Label class 15 exceeds nc=5 in data/custom.yaml. Possible class labels are 0-4
qt.qpa.xcb: could not connect to display qt.qpa.plugin: Could not load the Qt platform plugin "xcb" in "/home/ec2-user/anaconda3/envs/yolov7_custom/lib/python3.9/site-packages/cv2/qt/plugins" even though it was found. What should be done for this error?
Looks like you are running the script on a system with only CLI (without GUI). You would not be able to use functions that use the graphical output (such as cv2.imshow()).
Can I use labelme instead of labelImg? Whenever I use labelImg rectbox, it automatically terminates. Also whenever I install pip, the new directory 'labels' does not appear
Hello.. I am getting following error.. while installing pip install -r requirements_gpu.txt ERROR: Could not find a version that satisfies the requirement torch==1.11.0+cu113 (from versions: 1.7.1, 1.8.0, 1.8.1, 1.9.0, 1.9.1, 1.10.0, 1.10.1, 1.10.2, 1.11.0, 1.12.0, 1.12.1, 1.13.0, 1.13.1) ERROR: No matching distribution found for torch==1.11.0+cu113 Any suggestion?
Brother please help If the program detects i want an alarm But I don't know what to write i 'if' condition (what will be name of the detected image from the video)
if we run a webcam on this are we able to log detections and if for example class x detected then take screenshot and check for class y - if class y found crop - enhance image and then run ocr and then log result in a database?
It says: ERROR: Could not find a version that satisfies the requirement torch==1.11.0+cu113 (from versions: none) ERROR: No matching distribution found for torch==1.11.0+cu113 when i try to install requirements_gpu.txt
Hey Man, thanks for the great vid! I'm having issues with my mAP values, they are extremely low. I'm training on a dataset with 70 images and labels. I've run the training model a few times from 100 epochs to 1000 epochs yet still my recall (0.1-0.5) and mAP graphs are very low (between 0.001 and 0.0015). No bounding boxes appear on the test image (on custom object detector run) unless I use an extremely low confidence (0.001). Although a bounding box does appear, it just covers the entire image (not the requested custom object). Any ideas what I could do to solve this issue? I would highly appreciate it, Thanks.
These many images work for one object. But if there are multiple objects, and that too complex ones, 70 images are too few. You can try training yolov7 tiny and see if it improves mAP.
@@TheCodingBug Thanks for the quick reply! I'm training a relatively simple single object. I'll run a training set on my data using the yolov7 tiny model and see if that helps. Thanks for the suggestion my friend!
I tried training for tomatoes and tomato leafs but all the bounding boxes came up labeled as tomatoes. Is there a way to make it so leafs are detected as leafs and tomatoes are detected as tomatoes? I labeled them differently in labelImg so i dont know how they came up with same label after running detection. @TheCodingBug
Thanks for the amazing tutorial; it was incredibly helpful! I am wondering how to go about training my custom detector beginning with pretained weights from coco or imagenet, instead of training from scratch. Is there some way that I can modify the existing code in your tutorial to do this?
When we mention the model weights (.pt) file in the command, basically we are using pretrained weights from coco dataset and then fine-tune on our data.
File "C:\capstone\yolov7_custom\yolov7-custom\train.py", line 592, in device = select_device(opt.device, batch_size=opt.batch_size) File "C:\capstone\yolov7_custom\yolov7-custom\utils\torch_utils.py", line 71, in select_device assert torch.cuda.is_available(), f'CUDA unavailable, invalid device {device} requested' # check availability AssertionError: CUDA unavailable, invalid device 0 requested What should be done for this error?
@@dhruvinkanani670 could you help me, I get error Cum_counts = np.cumsum(no.greater(counts,0,dtype=np.int64)) TypeError: No loop matching the specified signature and casting was found for ufunc greater.
please if i would use the weights of the first training and run the model another time with more data should i modify in the commande the weight yolov7.pt with best.pt of the privious model and that's it ?
Great tutorial, but when I start training I get this error: (ModuleNotFoundError: No module named 'numpy') I tried installing numpy with (pip install numpy) but it tells me (Requirement already satisfied: numpy), I did everything you did properly but how can I fix this?
I use colab to train it, but it's not good. my mAP@.5:.95: just "0.3x" i follow your step, epochs 100 batch_size 8 should i train more epochs? or add more images?
@@TheCodingBug ok thank you!! I watch in 10:27 you just add less than 10 images in training, your mAP@.5:.95: have 0.6x in 10:47 I have a little confuse, how much images need to add.
Thanks for making this video. I am facing problem in installing "torch==1.11.0+cu113" due to restriction i am not able to solve it , getting SSL error. In my system torch==1.13.0+cu113 is already present but its not working for the code. My model get train for 3 epoch than it shows the "Nvidia driver error". Please suggest something i would be grateful to you.
At the final step of running the custom object detector part I get an error saying that 1.jpeg does not exist when I am positive that I have it in the yolov7-custom folder. Any suggestions pls
@@TheCodingBug Traceback (most recent call last): File "C:\Users\alexa\OneDrive\Documents\Comp Sci\YOLOv7_Custom\yolov7-custom\detect.py", line 195, in detect() File "C:\Users\alexa\OneDrive\Documents\Comp Sci\YOLOv7_Custom\yolov7-custom\detect.py", line 57, in detect dataset = LoadImages(source, img_size=imgsz, stride=stride) File "C:\Users\alexa\OneDrive\Documents\Comp Sci\YOLOv7_Custom\yolov7-custom\utils\datasets.py", line 138, in __init__ raise Exception(f'ERROR: {p} does not exist') Exception: ERROR: C:\Users\alexa\OneDrive\Documents\Comp Sci\YOLOv7_Custom\yolov7-custom\1.jpeg does not exist This is the error message that I get
getting an error when trying to train about no labels existing, even though the images were labelled and are in the correct dir, tried searching online and coming up at a loss - scanning images and labels... 0 found, 126 missing, 0 empty, 0 corrupted: 100% - AssertionError: train: no labels in data\train\Images.cache. Can not train without labels.
hello, i have a question why is my mAP very low like the accuracy is only 0.2-0.3 did i miss any modifications? ive ran 3 training already but the accuracy is very bad :(
@@TheCodingBug Hello I was trying to add a secondary object. I changed the number of objects to 2 in CFG/training/yolov7-custom and data/(coco)custom_data, i also added another object on the list. train.py: error: argument --weights: expected one argument Could you please help me to fix this 🙏
Any suggestions for getting my mAP values up? Almost all my mAP 0.5s are very low numbers for each epoch and it doesn't detect the jack sparrow face even at --conf 0.1
I was having the same issue. Go to runs > train folder and look at the graphs. In my case , I was not getting the box because it wasn’t learning. You may have to add more pictures and labels.
@@alejandroguerra6755 That's exactly why. I am running on a low end GPU so I didn't train it long. Lowering confidence parameter from 0.5 to 0.1 did produce the box.
Thanks a lot for a very effective, fast and complete, tutorial, not wasting people time.
Can't agree more😀
Thanks a lot. I have been trying to train a model for a week and finally I found this gem. I was having an issue regarding numpy: AttributeError: module 'numpy' has no attribute 'int'. The problem was with the latest version of numpy. I ran "pip install --upgrade numpy==1.23.5" and it fixed the issue.
Thank you, dude. I have already completed my thesis using the algorithm. God Bless U
Thank you very much!Everything works perfectly (except for a few issues here and there,but it was all solved at the end),i really appreciate this!
hello, I require assistance because when I run detect it doesnt detect jack sparrow's face, and my highest accuracy was only 0.78 :(
@@khenpahilanga9596 I am having the same issue....we're you able to resolve it?
I followed the steps that you mentioned in the video and I tried 4-5 times to train the model but in the end, after training it did not detect Jack Sparrow's face in the image when trying to give an input image
Man keep these simple efficient videos about SOTA CV models, thanks alot ❤
Have not tried it yet but its the best tutorial I could find. Very good content!
what a nice tutorial! This video makes me surprised and more easier to realize the mechanism of yolov7.
Great video! no time wasting. Can you please tell me how can I train coco format labeling data for instant segmentation?
I am planning the video for the next week.
Hello, thanks for this video, straight to the point and no time wasted, I just wanna ask after the first training of custom dataset. how do i add more dataset unto the first trained custom dataset yolo? do i use the best weights if i wanna continue adding more dataset to improve accuracy and more classes? becuz i wanna train my yolo while it retains the previous data is was trained with without starting from scratch, so i was wondering if u could help me with that. Thank you alot :)
Thanks for this great and clear tuto. Can you make another video on yolov7 object detection in real time using webcam?
Just use --source 0 for webcam along with --view-img
@@TheCodingBug thank you so much you are the best!!
Thank you so much, I’m doing my final year project and this video helps me so much.
But i have a question. If i trained it to detect 1 object after that i want to train it to detect another object but still keep the previous one. How can i do that?
Here's their TH-cam channel with tutorials on how to use it: th-cam.com/channels/5gLr7dLtV3Xe0eJvHOImNA.html
Hello. I run it on laptop with RTX 3060, 6Gb vram. But it runs only if batch-size is 1. Otherwise it doesn't work, an error occurs: RuntimeError: CUDA out of memory. Tried to allocate 1.92 GiB (GPU 0; 6.00 GiB total capacity; 545.16 MiB already allocated; 3.80 GiB free; 588.00 MiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF. I use 2000 images for training.
same here
Thank you so much! What does workers = 1 mean in the training command? Appreciate your help
Number of workers for data loading.
Thank you! Helped me a lot in my school project :D Amazing video :D
Hi Sir, as told you in the recent past, thank you so much for your effort producing this awesome minicourses. I'd like so much if you could produce a Yolov7 or Yolov8 detect video for Raspberry Pi. What's your thought regarding this idea?
Great video...!
I followed the same steps as in the video but ended with no detection - I have trained 10 images and labels. May I know what missing in the detection?
Add around 100 images. 10 are too few.
i trained with 500 images but getting no detection(code runs properly but it detects nothing)
@@ujjwalkatiyar9998 try training and testing on Google colab. Perhaps it'll solve the problem. th-cam.com/video/_fXABNYlZhY/w-d-xo.html
@@ujjwalkatiyar9998 does it solve the issue please ?
What's your PC specs? I wanna know just to get a rough idea of compute time in training
I am using GTX 1060 Laptop with coreI7 and 16GB DDR4 RAM.
@@TheCodingBug oh, well the ram information is less relevant.
What's the VRAM on the 1060?
It would seem my 3060 desktop gpu is an ideal choice as I can train yolo on it in a relatively shorter time!
@TragicGFuel 1060 is 6GB.
3060 is faster, probably 2x.
@@TheCodingBug after some reasearch it seems that the VRAM being 12GB is a real boost to the training.
The fact that I can overclock it is also a bonus.
Excellent video. Is there an easy way to estimate the distance to detected vehicles?
If you have a stero camera you can measure detected object distance. Yolo is giving coordinates of objects. Stero camera can use this coordinates and can take distance. We are working on autonoums car and we are detecting traffic sign and their distance for writing best algorithms.
Please make a video on how to custom train already custom trained model
for example : I trained model for detecting trash and wants to improve the accuracy, so I again want to train the model with different dataset
Hello Sir, have you planned to do similar videos regarding the new Yolov8? It should be very interesting
Very clear! Such a nice video :)
I just have a few questions...
Would it be possible to do the inference with the pretrained weights (for instance, base yolov7) detecting only one class (for instance, person; or car)?
In case it is, could yo please explain how?
Regards
Yes it's possible. Follow pretrained yolo v7 tutorial (if you haven't already) and use flag
--classes 0
to detect person class only. Detail about this flag is discussed here th-cam.com/video/K_OGcfXwskc/w-d-xo.html
Thank you so much! Now I am facing another problem...
For my custom dataset, my images are non squared, they are 1280x720. I saw a flag "--rect" in order to show I am not using squared images for training. How should I define the "--img-size" flag? I don't think it is just typing 1280 720; but I don't find anything relevant surfing on the Internet.
LabelImg only create boxes, it is possible to do Instance segmentation in this Yolo ? What label-software to use?
If I want the process to go faster when training yolov7 what should I do with workers
My pc is not having cuda availability. Can I have any alternative... Like running same in Google colab??
Yes you can. Yolov7 on colab tutorial is also there on my channel.
Hi there @TheCodingBug! First of all thank you for this Video!
I have two Questions:
1. Can you maybe make a deep dive Video for this? I mean i could perform the custom training. But it still just runs in cmd-window. Can you maybe make a detailed guide on the main-function and which parts do what? . For example which lines to modifiy in the code if i want to for example modifiy the output of yolo? In v3 i wrote a function that only outputs sth if a person is detected , but no text output for anything else. How to do this in yolov7?
2. Also im highly interested in transfer learning. In this video we made a custom model with just one class. But how can i for example add new custom detections to the already existing coco-model ?
Hi,
I am having this error
RuntimeError: CUDA error: no kernel image is available for execution on the device
Whereas my cuda and pytorch versions are the same
Name: torch
Version: 1.11.0+cu113
I had 11.6 but i changed it to 11.3 now it should work but its not and iam stuck now.
EDIT: I have gtx 770 which is old. But i want to run yolo can i run only on cpu ? As i have installed only cpu pytorch but it gave this error:
AssertionError: CUDA unavailable, invalid device -1 requested
Please Assist ASAP
Could you please explain the steps needed to train more on top of initial training?
What steps to take after doing all the steps in this video(ways to move through yolov7-custom folder/anaconda prompt)?
What files need to be deleted/altered?
Generally, what do i need to do to train again to add more classes/img’s/annotations?
Thank you very much.
hello, have u gotten an answer regarding your question? I would also like to know
I got no bounding box on the result image, is it because I used only 3 images for training?
Most likely yes.
Hi, this is a very nice video man! I have a question, is it possible to convert this model to TFLite following the same steps that you did on YOLO v4 video? Thanks a lot!
No. That only works for tensorflow 2.3.1.
@@TheCodingBug Okay, thanks
how to create a empty script as u mentioned in 1:20......?
Create a new text file, rename it with extension .py instead of .txt
I am using same method but show wrong output like above the image show clock to predicted every image . How to solve the problem?
how can we put this in a .py file so that we can use webcam support and acctually do stuff with the object detection
This is already a .py file!
If you want to use webcam, just use --source 0 instead of video path.
@@TheCodingBug Namespace(weights='yolov7.pt', cfg='cfg/training/yolov7-custom.yaml', data='data/custom-data.yaml', hyp='data/hyp.scratch.custom.yaml', epochs=100, batch_size=8, img_size=[640, 640], rect=False, resume=False, nosave=False, notest=False, noautoanchor=False, evolve=False, bucket='', cache_images=False, image_weights=False, device='0', multi_scale=False, single_cls=False, adam=False, sync_bn=False, local_rank=-1, workers=1, project='runs/train', entity=None, name='yolov7-custom', exist_ok=False, quad=False, linear_lr=False, label_smoothing=0.0, upload_dataset=False, bbox_interval=-1, save_period=-1, artifact_alias='latest', freeze=[0], v5_metric=False, world_size=1, global_rank=-1, save_dir='runs\\train\\yolov7-custom4', total_batch_size=8)
tensorboard: Start with 'tensorboard --logdir runs/train', view at localhost:6006/
hyperparameters: lr0=0.01, lrf=0.1, momentum=0.937, weight_decay=0.0005, warmup_epochs=3.0, warmup_momentum=0.8, warmup_bias_lr=0.1, box=0.05, cls=0.3, cls_pw=1.0, obj=0.7, obj_pw=1.0, iou_t=0.2, anchor_t=4.0, fl_gamma=0.0, hsv_h=0.015, hsv_s=0.7, hsv_v=0.4, degrees=0.0, translate=0.2, scale=0.5, shear=0.0, perspective=0.0, flipud=0.0, fliplr=0.5, mosaic=1.0, mixup=0.0, copy_paste=0.0, paste_in=0.0, loss_ota=1
wandb: Install Weights & Biases for YOLOR logging with 'pip install wandb' (recommended)
fatal: not a git repository (or any of the parent directories): .git
Traceback (most recent call last):
File "C:\Users\Arsh\Desktop\arshaim\yolov7-custom\utils\google_utils.py", line 26, in attempt_download
assets = [x['name'] for x in response['assets']] # release assets
KeyError: 'assets'
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File "C:\Users\Arsh\Desktop\arshaim\yolov7-custom\train.py", line 616, in
train(hyp, opt, device, tb_writer)
File "C:\Users\Arsh\Desktop\arshaim\yolov7-custom\train.py", line 86, in train
attempt_download(weights) # download if not found locally
File "C:\Users\Arsh\Desktop\arshaim\yolov7-custom\utils\google_utils.py", line 31, in attempt_download
tag = subprocess.check_output('git tag', shell=True).decode().split()[-1]
File "C:\ProgramData\Anaconda3\envs\yolov7\lib\subprocess.py", line 424, in check_output
return run(*popenargs, stdout=PIPE, timeout=timeout, check=True,
File "C:\ProgramData\Anaconda3\envs\yolov7\lib\subprocess.py", line 528, in run
raise CalledProcessError(retcode, process.args,
subprocess.CalledProcessError: Command 'git tag' returned non-zero exit status 128.
hello great video. Question: how the official YOLOv7 wants the bounding box format to be? Just like COCO format [xmin, ymin, width, height] or like previous YOLOs [xcenter, ycenter, width, height]?
Thank you for this amazing video. But, I am having some problems. When I run detect.py, no bounding box appears. Where is the problem coming from? I have Detectron and OpenCV installed. Are these necessary to visualize bounding boxes? I hope you can help me.
Detectron is not necessary. Opencv is.
With regards to no bounding box, check precision and recall graphs in training folder (where best.pt file is located). If it has really small values, the model is struggling to learn and you need more data for training.
@@TheCodingBug Actually I wanted to test the detection with pre-trained weights by calling "python detect.py --weights yolov7.pt --source inference/images/horses.jpg --img 640". But, no bounding boxes appears in horses.jpg.
I have approximately 2000 images for 4 classes. I trained the model in YOLOv5, thanks to your videos. I got 0.65-0.70 mAP. I trained the model in YOLOv7, but precision and recall stayed under 0.001 too low.
I guess the problem comes from former issue. Is there any idea to solve this?
@@tugbayildizli3828 Check the other video for pretrained model. th-cam.com/video/n2mupnfIuFY/w-d-xo.html
See if it works properly. If not, try updating Nvidia drivers for GPU usage or try CPU device instead of GPU.
@@TheCodingBug Thank you for your response. I used CUDA 11.3. I installed CUDA 10.2 and the problem solved.
@@TheCodingBug Hi SIr, I'm on the same situation. What does it means "...you need more data for traing"? Maybe more images (I have 57 images on Train folder). Thanks in advance
Hi there... can we have a conversation about object detection? would love to show you what we are building and to understand how OD can boost our project.
Hello. Let's connect on Facebook. facebook.com/BugCodingThe
your channel is gold mine
I am glad you found it helpful.
You are a hero.
i have doubt like cant we run the same thing in intel laptops because i got an error regarding cuda???
You can run without cuda on CPU.
@@TheCodingBug but isn't it like the project is completely based on cuda ??
@@indstorm8680 No. It needs cuda and Nvidia GPU for faster training but cpu can also be used for slower training.
Hello brother , can you please let me know what is the software you used to modify the official files ? At 7.25 .
Please reply soon , I need it urgently for my University Project
Sublime text.
Amazing my friend thank you for posting this
hello sir. thank you for the explanation. i try to deploy with my 30 minutes video duration but turns out the detection video becomes 16 minutes. does it related to fps?
great one appreciate your efforts, thanks a lot
Hi. I’m having an error while training.
AssertionError: Label class 15 exceeds no=1 in data/custom_data.yaml. Possible class labels are 0-0.
Not sure how to solve this.
i am also having the same issue how to resolve?
I couldn't install torchvision and torch. I do the same thing but I got an error. How can I fix this issue?
Hello, I am trying to train a model on custom plant images the model gets trained but does not detect correctly and makes no bounding box, is it because of underfitting, because I have only 60 images of a plant specie, is there any way to train the model with low number of images
how can I do this training faster using my computer's GPU.Can you help me?
I get this error when i try to train
No loop matching the specified signature and casting was found for ufunc greater
Can I please know the reason?
Great video. Thanks
Thanks for sharing your knowledge
Great tutorial!
hello could you please help me, i train dataset but train also show 0 label in the box. and when i try to test, there is no bounding box label. whats wrong with my dataset ?
I am getting AssertionError: Label class 12 exceeds nc=1 in data/custom-data.yaml. Possible class labels are 0-0
You have specified number of classes as 1 but have more classes in labels
Hello good sir, I'm trying to train multiple objects at once but I'm getting "AssertionError: Label class 2 exceeds nc=2 in data/custom_data.yaml. Possible class labels are 0-1". I've already change number of classes to 2. What have went wrong here?
i am having the same issue how did you resolve?
Excuse me. Do this work with live camera detection?
Just use --source 0 instead of video file.
hi, can i ask yuo which yolov8 version should I choose to train the model on a custom dataset? (yolov8m, yolov8l, ...)
Hi, nice video. Really Helpful. But I have a problem when training. I have this error:
MP: Error #15: Initializing libiomp5md.dll, but found libiomp5md.dll already initialized.
I searched around and they say the solution is to install all packages without MKL support:
conda install nomkl
is this a good solution? I mean is it because my CPU is old? It is Intel(R) Core(TM) i7-7700 CPU
I am not sure. Never encountered this error.
@@TheCodingBug Oh ok, I will try it on a new comp and see what happen. Thanks for the reply.
Just an update in case someone had the same issue as me. I found out what the problem was. It is because of torchaudio package which was installed on my env. I noticed that in this video you remove torchaudio from the installation whereas in one of your other first videos you just copied the line from pytorch website which has torchaudio in it. I basically remake the env without torchaudio and everything works fine. Thanks again for the well made tutorial.
@@stephencolbert934 I am glad you were able to solve the problem.
@@stephencolbert934 hello can u give more info? I have the same problem and i dont know what i should change... ;/
at 3:03 I get the error " name 'labelImg' is not defined". I installed the labelImg fine but when I want to open it by writing "labelImg" its doesn't work.
I am getting this type of error what could be the reason
assert mlc < nc, 'Label class %g exceeds nc=%g in %s. Possible class labels are 0-%g' % (mlc, nc, opt.data, nc - 1)
AssertionError: Label class 15 exceeds nc=5 in data/custom.yaml. Possible class labels are 0-4
In your annotations, you have class label 15 but you've mentioned number of classes = 4.
@@TheCodingBug no i have five classes and same number of classes i have mention in cfg and coco file
Is there a way in Yolo not just to Annotate , but also to remove Background ? Where i can find?
No it's not possible without semantic segmentation.
@@TheCodingBug OK, thank you👍🏻, but which repository should I use ?
qt.qpa.xcb: could not connect to display
qt.qpa.plugin: Could not load the Qt platform plugin "xcb" in "/home/ec2-user/anaconda3/envs/yolov7_custom/lib/python3.9/site-packages/cv2/qt/plugins" even though it was found.
What should be done for this error?
Looks like you are running the script on a system with only CLI (without GUI). You would not be able to use functions that use the graphical output (such as cv2.imshow()).
@@TheCodingBug i'm using SageMaker terminal in AWS, what i can do in this case?
Awesome, thank you!
Can we do this in tensorflow, because i already have cuda 11.2 and cudnn 8.1. But i am not getting gpu support for pytorch
Can I use labelme instead of labelImg? Whenever I use labelImg rectbox, it automatically terminates. Also whenever I install pip, the new directory 'labels' does not appear
Use label studio instead
How to install label studio
@@ArunkumarN-dg1fz pip install label-studio
Then run it using command
label-studio
awasome, thank you
Hello.. I am getting following error..
while installing pip install -r requirements_gpu.txt
ERROR: Could not find a version that satisfies the requirement torch==1.11.0+cu113 (from versions: 1.7.1, 1.8.0, 1.8.1, 1.9.0, 1.9.1, 1.10.0, 1.10.1, 1.10.2, 1.11.0, 1.12.0, 1.12.1, 1.13.0, 1.13.1)
ERROR: No matching distribution found for torch==1.11.0+cu113
Any suggestion?
See how I created requirements_gpu.txt ... Follow that to create for new version.
@@TheCodingBug it worked.. :)
@@MayankDayal1234 what version did you go with please tell?
@MayankDayal1234 i am also getting same error plz say how did you debug it
Brother please help
If the program detects i want an alarm
But I don't know what to write i 'if' condition (what will be name of the detected image from the video)
Thanks for the video!
I followed this tutorial but after training, there are no bounding boxes appearing in the image. Any idea why is that?
I am having the same issue......are you able to solve it?
if we run a webcam on this are we able to log detections and if for example class x detected then take screenshot and check for class y - if class y found crop - enhance image and then run ocr and then log result in a database?
No. You will have to modify the code to implement these specific functionalities.
It says: ERROR: Could not find a version that satisfies the requirement torch==1.11.0+cu113 (from versions: none)
ERROR: No matching distribution found for torch==1.11.0+cu113
when i try to install requirements_gpu.txt
Install latest version of pytorch gpu with the command taken directly from official website.
But won't it exceed the required Version?
If not pls give the line I needed
You'd be able to find previous versions there too. But in my experience, latest version is still compatible.
@@TheCodingBug thank you
Hi sir, I wanna ask, can I recognized someone with this? Or just detection?
You can recognize people.
@@TheCodingBug thank you so much sir💙
Hey Man, thanks for the great vid!
I'm having issues with my mAP values, they are extremely low. I'm training on a dataset with 70 images and labels. I've run the training model a few times from 100 epochs to 1000 epochs yet still my recall (0.1-0.5) and mAP graphs are very low (between 0.001 and 0.0015). No bounding boxes appear on the test image (on custom object detector run) unless I use an extremely low confidence (0.001). Although a bounding box does appear, it just covers the entire image (not the requested custom object).
Any ideas what I could do to solve this issue? I would highly appreciate it, Thanks.
These many images work for one object. But if there are multiple objects, and that too complex ones, 70 images are too few.
You can try training yolov7 tiny and see if it improves mAP.
@@TheCodingBug Thanks for the quick reply! I'm training a relatively simple single object. I'll run a training set on my data using the yolov7 tiny model and see if that helps. Thanks for the suggestion my friend!
@@TheCodingBug can you answer please how to select the best number of images to train with
@@chaimacham9414 100 images for 1 object is a good starting point.
thank you so much for your video it is really really helpful
Holy shit this was amazing
I tried training for tomatoes and tomato leafs but all the bounding boxes came up labeled as tomatoes. Is there a way to make it so leafs are detected as leafs and tomatoes are detected as tomatoes? I labeled them differently in labelImg so i dont know how they came up with same label after running detection. @TheCodingBug
Thanks for the amazing tutorial; it was incredibly helpful! I am wondering how to go about training my custom detector beginning with pretained weights from coco or imagenet, instead of training from scratch. Is there some way that I can modify the existing code in your tutorial to do this?
When we mention the model weights (.pt) file in the command, basically we are using pretrained weights from coco dataset and then fine-tune on our data.
Ah I see. Awesome, thanks!
My lebels are generated by labelImg, but why the system log corrupted: 100%, and it can not go traing....
how if i want to change thickness bonding box and size of font when predict object
Thank you so much, You solve my problem with your video. I wish you could do also real-time detection with camera :(
Use --source 0 instead of video file for camera.
Hello, i use 457 images for training and 196 for validation, but i have MAP very low 0.064!! please any recommandation?
What's the object?
I sent u à message on patreon.
may i know your device specs, including cuda version and nvidia driver version? thanks
File "C:\capstone\yolov7_custom\yolov7-custom\train.py", line 592, in
device = select_device(opt.device, batch_size=opt.batch_size)
File "C:\capstone\yolov7_custom\yolov7-custom\utils\torch_utils.py", line 71, in select_device
assert torch.cuda.is_available(), f'CUDA unavailable, invalid device {device} requested' # check availability
AssertionError: CUDA unavailable, invalid device 0 requested
What should be done for this error?
You do not have cuda installed (or do not have nvidia GPU).
If you have nvidia GPU, update the drivers and install pytorch with cuda support.
Remove -device 0 from command line argument when you run the program
you have to recreate a new environment and then redownload pytorch in that environment and then try the training cmd. it works for me.
@@dhruvinkanani670 could you help me, I get error
Cum_counts = np.cumsum(no.greater(counts,0,dtype=np.int64))
TypeError: No loop matching the specified signature and casting was found for ufunc greater.
@@isnaayu_m could you fix this error?
please if i would use the weights of the first training and run the model another time with more data should i modify in the commande the weight yolov7.pt with best.pt of the privious model and that's it ?
Yes. Also, delete labels.cache files in train and val folders before starting training on new data.
@@TheCodingBug okk thank you so muchh
Great tutorial, but when I start training I get this error:
(ModuleNotFoundError: No module named 'numpy')
I tried installing numpy with (pip install numpy) but it tells me (Requirement already satisfied: numpy), I did everything you did properly but how can I fix this?
How much gpu memory is required to run on local system. Is 2GB is enough to train with 1000 images.
Maybe if you use batch size of 2 or 4, it would work on 2GB GPU.
ModuleNotFoundError: no module named ‘numpy’
I have installed numpy plz help this is occurring on the python train command section
I use colab to train it, but it's not good. my mAP@.5:.95: just "0.3x" i follow your step, epochs 100 batch_size 8
should i train more epochs? or add more images?
Add more images.
@@TheCodingBug ok thank you!! I watch in 10:27 you just add less than 10 images in training, your mAP@.5:.95: have 0.6x in 10:47
I have a little confuse, how much images need to add.
@@remiel0731 I have 10 images in validation, for training, I have 50+ images.
@@TheCodingBug Sorry to bother you. I'm realize than I just need to give more images to help it make great that's all, don't think to much....😂😂
How I can reach you for further learning and seek your support for my Ph. D Sir.
Thanks for making this video. I am facing problem in installing "torch==1.11.0+cu113" due to restriction i am not able to solve it , getting SSL error. In my system torch==1.13.0+cu113 is already present but its not working for the code. My model get train for 3 epoch than it shows the "Nvidia driver error". Please suggest something i would be grateful to you.
Just update the driver.
@@TheCodingBug Thanks for replying.
I already have “Nvidia driver” installed.
Is there any requirements of specific torch versions I.e 1.11 or can I use the existing toch of 1.13 version. Which is already available
@@ravichoudhary2365 Install the latest version.
@@ravichoudhary2365 Check 5:09
How to train model for multiple classes??
th-cam.com/video/gRAyOPjQ9_s/w-d-xo.html
Follow this method to create multi class dataset and train yolov7 or yolov8
At the final step of running the custom object detector part I get an error saying that 1.jpeg does not exist when I am positive that I have it in the yolov7-custom folder. Any suggestions pls
Check if it's extension really is jpeg.
@@TheCodingBug
Traceback (most recent call last):
File "C:\Users\alexa\OneDrive\Documents\Comp Sci\YOLOv7_Custom\yolov7-custom\detect.py", line 195, in
detect()
File "C:\Users\alexa\OneDrive\Documents\Comp Sci\YOLOv7_Custom\yolov7-custom\detect.py", line 57, in detect
dataset = LoadImages(source, img_size=imgsz, stride=stride)
File "C:\Users\alexa\OneDrive\Documents\Comp Sci\YOLOv7_Custom\yolov7-custom\utils\datasets.py", line 138, in __init__
raise Exception(f'ERROR: {p} does not exist')
Exception: ERROR: C:\Users\alexa\OneDrive\Documents\Comp Sci\YOLOv7_Custom\yolov7-custom\1.jpeg does not exist
This is the error message that I get
@@TheCodingBug I checked if the extension was jpeg and it was
I solved the problem that change path(english+my own language -> only english)
No labels in data\train\imagenew.cache. Can not train without labels. I am getting this errror even though I followed all the above steps
Your path to train is wrong. I never used imagenew anywhere.
getting an error when trying to train about no labels existing, even though the images were labelled and are in the correct dir, tried searching online and coming up at a loss - scanning images and labels... 0 found, 126 missing, 0 empty, 0 corrupted: 100% - AssertionError: train: no labels in data\train\Images.cache. Can not train without labels.
Your labels in "data\train\labels" folder must be in txt format. Check if they are in .xml format and change them to .txt format.
At the time of testing it can not detect any picture and video
hello, i have a question why is my mAP very low like the accuracy is only 0.2-0.3 did i miss any modifications? ive ran 3 training already but the accuracy is very bad :(
Did you solve it?
Hi ! Do you know if cuda is needed for this script ?
It's not needed but desirable as training without cuda will be very very slow.
@@TheCodingBug
Hello
I was trying to add a secondary object.
I changed the number of objects to 2 in CFG/training/yolov7-custom and data/(coco)custom_data, i also added another object on the list.
train.py: error: argument --weights: expected one argument
Could you please help me to fix this 🙏
Any suggestions for getting my mAP values up? Almost all my mAP 0.5s are very low numbers for each epoch and it doesn't detect the jack sparrow face even at --conf 0.1
Hey, could you solve it??
another great video. tks u
requirment_gpu.txt is not found sir what should i do now
Thats because you skipped the video :)
I created it myself. So you need to do it too. Watch: (5:50)
when I run the detect.py, its not producing the bounding box around the object.
I was having the same issue.
Go to runs > train folder and look at the graphs.
In my case , I was not getting the box because it wasn’t learning.
You may have to add more pictures and labels.
@@alejandroguerra6755 That's exactly why. I am running on a low end GPU so I didn't train it long. Lowering confidence parameter from 0.5 to 0.1 did produce the box.