I started training dataset containing wheat images in colab and colab's runtime is just stopped at 1961 iteration inspite of setting setting javascript code in console. Can you please suggest any way we can prevent it so that training continues?
prateek gupta that can sometimes happen due to memory issues within the notebook after running for a long time. Just restart training from saved weights and you will be good as new!
When I am running the first step in step 5, it shows this. What should I do? CUDA Error: out of memory CUDA Error: out of memory: File exists darknet: ./src/utils.c:326: error: Assertion `0' failed.
I just saw that this video was posted. I was literally following your tutorials over the last few days to build a license plate detector for my University Major Project. You may have just saved my life, I'll be watching this video later
@@TheAIGuy An update on my project. Everything went well in my presentation besides a few questions on logistics. Thanks to you, it looks like I've passed my Major Project, the most important module to graduate from my university in Jamaica. All that's left to do is finish 2 modules to meet the credit requirements to graduate. Thank you for your video, it really helped me out of a tough spot
@@brandonchung5305 hello! Would you be open to sharing the code? I'm working on a similar project and i am a bit stuck. Or if possible, would you be open for a chat??
You teach better than paid courses which I've been taking for a couple of months. Please keep up the good work, thank you very much, you're teaching me a lot!
This is exciting, I was finally able to use your tutorial to train and test a working detector! you are making machine learning accessible to someone without a background in it, and I appreciate that. I only trained a small model (I did a beer detector lol), but this will be a great foundation for a larger paper I'm working on for my undergraduate final project. Thanks again!
Thanks, this tutorial still works as of 2/28/2022. The only issue is that Step 5 may not work on a free GPU instance. Google Colab offers plans to access High-RAM GPUs at $9.99/month. I bought the subscription, changed the runtime to a High-RAM GPU, and Step 5 worked for me.
You are a life saviour! I have a senior project and had to train a model on custom data, your notebook and explanation literally saved me! Fr! I learned from you more than 3 college courses of ai, thank you so so much
Hi! First of all I want to thank you for all your videos and the way you share your knowledge, that is invaluable and appreciated! A while ago, I trained a single class with YOLOv3 to detect Argentine license plates (the old ones and the new ones) with 340 images. Of the 340 images you can detect 339, there is always one that does not detect it, it does not matter. I have done it because I wanted to try a Raspberry Pi and a webcam, do license plate recognition to recognize the characters and save them in a database, and that's where I am limited with my knowledge. Once I recognize where the patent is and draw a frame around it (already done), I need to extract it, correct its perspective, and do the OCR. I have seen that there exists YOLO, Tiny YOLO, Tensor Flow, Kereas, etc, and all of them using OpenCV; and the truth is that I already got lost. In your experience, what would be the ideal path to follow for a modest project like the one I want to do? Thank you very much.
I would recommend OpenCV with python. You can save the YOLOv4 detections and then run the OCR on the saved image. I will try and create a video showing how to do this on my license plate detector.
@@TheAIGuy Please make video on extracting licence plate only from this image and applying OCR pytesseract on same..also for video as well.. please sir
I would like to thank you for this video and the collab notebook you single handedly saved my final year project. And i hope you are doing well man. Keep up the good work you are the best
Thanks bro .Your videos are just amazing .I just want to suggest that kindly make video regarding test data on custom trained model and also a video regarding description of the output of yolo v4.Other than that you are just amazing
This video has helped me a lot :D. Congratulations on the tutorial!! Will you make a video on how to build an Object Tracker using YOLOv4? I think it would be interesting and would help a lot of people to develop different applications.
I got follwing error while I ran !python \convert_annotations.py command error is Currently in subdirectory: validation Converting annotations for class: Vehicle registration plate 0% 0/30 [00:00
I need to thank you for your job, it's obvious that this didn't worked everything as expected and I ended up solving many problems all along the training, but that's how we make science, discovering and understanding; Of course that without you it would be a LOT harder, thank you, I finally trained my first model!
As usual a really helpful tutorial, just wanted a suggestion how to store all the detected object co-ordinates from the video in a Jason file for further processing.
I try to do it with my custom dataset with 38 classes and always get "CUDA status Error: file: ./src/dark_cuda.c : () : line: 373 : build time: Sep 22 2020 - 09:47:07 CUDA Error: out of memory CUDA Error: out of memory: File exists darknet: ./src/utils.c:325: error: Assertion `0' failed.", what can be the reason? I've already set up division for 64 and batch for 2 images and random to 0 etc.
Fantastic you nearly done my entire final year project😁, would you do YOLOv4 and Deep SORT for multi-object tracking in the future? I know you did YOLOv3 and Deep SORT and you nailed it.
I did training on yolov4 on google colabs. and I only get last weight no backup files per 1000 iterations. How can be like that? fyi, I did training for 36 classes, and I don't think there's anything wrong with my backup file settings. is it okay? or i made mistake? thank u before!
Thanks a lot for this amazing tutorial, I really appreciate your efforts in preparing this wonderful video and this detailed notebook, god bless you :)
I have followed all the steps and mounted the drive properly hence gdrive appears on the file manager tab. However, while training I have followed the mydrive/yolov4/backup in both obj.data and created the folders in the drive but the trained weights are not saved. If I load a weights file/pre trained weights it can load it properly and resume training but it does not save the weights after 100 iterations.
You would want to edit yolov4-tiny.cfg inside the cfg/folder for the custom cfg. As well you will need to download the initial tiny weights for training instead of the .conv file used for full yolov4.
This video has saved my life! Much respect, thanks a lot for all the time and effort you put into your videos. Ive trained a model to identify a pallet of boxes. I have a question, would it be possible to have a video and track an object, in my case a pallet, and then perform a some edge detection or detect another class of objects ONLY within the bounding box of the identified pallet? Basically I'm wanting to count the number of boxes on the pallet and I have it working with images, but video is the next step. Any advice would be awesome.
@The AI Guy Thanks a lot for this video. I have trained the model on Custom Dataset of Retail Receipt Detection. It's working fine. But I think my model is over fitted. I have early stop the training at 1100 iteration. This is the MAP of my checkpoint. Done! Loaded 162 layers from weights-file calculation mAP (mean average precision)... Detection layer: 139 - type = 27 Detection layer: 150 - type = 27 Detection layer: 161 - type = 27 152 detections_count = 172, unique_truth_count = 150 class_id = 0, name = retailReceipt, ap = 99.09% (TP = 149, FP = 2) for conf_thresh = 0.25, precision = 0.99, recall = 0.99, F1-score = 0.99 for conf_thresh = 0.25, TP = 149, FP = 2, FN = 1, average IoU = 83.67 % IoU threshold = 50 %, used Area-Under-Curve for each unique Recall mean average precision (mAP@0.50) = 0.990915, or 99.09 % Total Detection Time: 8 Seconds
I have a question. I already follow your tutorial on building object tracking using yolov3 and it works. But how do I get some kind of result on the tracking in terms of value or graphs? I need to evaluate the tracker output. I hope you can help me with that. Thank you
Hi Bro, a cheers from HK! I hv followed every steps in yr tutorial, using a trained weight model (yolov3_8000) from others instead of yr pre-trained weights. When I directly do Step 7: Run Custom Object Detector without train it on Colab, it shows "Unable to init server: Could not connect: Connection refused " When I go to check the mAP of his model, it shows "Allocate additional workspace_size = 52.43 MB, Loading weights from /.weights...Couldn't open file: /.weights ". Can u advise how to solve this? Really really appreciate to u, bro!!!!
Thanks for the video, I learned alot from it. Just wondering how can I change so the prediction image doesn't only include border but it also shows the percentage of confidence.
yes check the *pyimagesearch* blog in which opencv has been used to read from the trained yolo model, subsequently taking the raw output and accepting only the detections related to some specific class.
After I start the training everything goes fine but when it prints out Detection layer: 161 - type = 28 Resizing, random_coef = 1.40 608 x 608 Create 6 permanent cpu-threads It stops. Please help. Thanks.
Check the train.txt and test.txt files (darknet > data > train.txt) if it's empty then something went wrong with the python scripts. I had the same problem because I manually labeled the images and used png images instead of jpg. So I had to replace jpg with png in the generate_train.py
I am training a detector for 'Persons'. My detection and validation datasets are 5000 and 1500. How many iterations will it take to bring the average loss down to 0.5 or mAP up to 95% ?
Hi am trying to train on desktop when am started training am getting the error message like could not load images even the paths are correct in train.txt any thoughts what maybe the issue ?
Hi, I could learn so many things in YOLO through your TH-cam! Now, I also want to see the tutorial on applying the model to the webcam. Do you have a plan to make tutorial about applying the model to the webcam? I hope so!
What if we wanted to just add 1 more class to the ones YOLOv4 can already detect? In particular I would like to detect both "people" and "faces" using the custom functions you made in a later video for detecting and cropping these 2 classes. - What changes would I need to make to classes.txt, obj.data and obj.names? - And to the variables __C.YOLO.CLASSES (coco.names) and __C.TRAIN.ANNOT_PATH (val2017.txt) of the config.py file of your custom functions? I lack some experience in this field and I'm not sure whether I should retrain the whole model to detect these 2 classes from scratch or if I can just retrain the model so that it can learn to detect faces while still being able to detect people as it can by default. Love your content, the most helpful and practical I've seen :)
Hi The AI Guy, Great video! I have a couple of questions. 1. Is there a way I can save the chart.png that shows both avg loss and map, to my google drive? 2. My weights saved every 10k iterations instead of 1k iterations, where would I make this change to make sure that the weights are saved every 1k iterations instead of 10k iterations? 3. Any recommendations/suggestions to keep in mind for OCR using YOLOv4?
Do we have to do any preprocessing to the images dataset augmentation before performing the steps or should we just start annotating the images and follow the steps from the video
Thanks for the video. When trying to unzip the files I get this error: Archive: ../test.zip warning [../test.zip]: 496177718 extra bytes at beginning or within zipfile (attempting to process anyway) error [../test.zip]: start of central directory not found; zipfile corrupt. (please check that you have transferred or created the zipfile in the appropriate BINARY mode and that you have compiled UnZip properly) (and same for obj.zip) Any ideas on what went wrong? I already made a few models following your tutorial for yolov3 in Cloud and never ran into this error.
I'm getting similar error: Wrong annotation: class_id = 101. My .txt all start with 101. Seems like more than just a coincidence. Not sure what I'm doing wrong or how to fix. Any idea Mr AI Guy? This video is thoroughly awesome.
@@brentdunklau4536 I think I solved this as my model seems to be training OK. Before you start labelling find the directory where LabelImg is installed on your computer. Inside that folder is a predefined_classes.txt. Delete all the entries inside this .txt file. What I think is happening is LabelImg comes with pre defined classes like dog, person, cat ect. There are 16 in all hence my wrong annotation class_id of 15 (LabelImg starts at 0). You have probably got 85 objects in your labelling but the model is looking for 101 including the dog, person, cat ect. I'm sure there's a way you can get around this without having to re-labelling all your data. Hope this works for you. Edit, this solved my problem.
One of the best videos on yolov4... Thanks for explaining it so well I have a small doubt... can we change the batch_size=2 and image_size=1024?? If we can do it what should be the value of subdivision??
When you retrain YOLOv4 this way, does it use any transfer learning? In other words, do the previous weights carry over to the new training and the new training does not retrain the model from scratch?
I get this error when i try to train with one class. I think there are places where the classes=1 to be set also some other changes. Can you share the .cfg file for training the above example with one class. Error: in the file data/obj.names number of names 1 that isn't equal to classes=80 in the file cfg/yolov4-obj.cfg
@@mattishamann8660 i am also facing that issue,while i have changed 3 classes which was =80 ,than =1 as per recommendation,how to get solve please inform
I wanted to add I was able to varify the train.txt file was in the darknet data folder on my VM, google colab. The first line reads >> data/obj/0145644b6378f2ea.jpg
Hello, I found this article to be excellent. May I inquire about the limitations regarding custom datasets? Specifically, I have prepared 65,000 labeled images along with corresponding text files. Would this dataset size be acceptable on google colab?
I have created my custom dataset with the YOLOv4 compatible annotations. I have a question to ask. 1. Where does the classes.txt file has to be placed?
I was trying to train yolov5 on custom data but when giving command for training , it is just exiting the cell after below message "Saving weights to backup/yolo-obj_final.weights Create 6 permanent cpu-threads If you want to train from the beginning, then use flag in the end of training command: -clear "
I am using multi class dataset in the generation of train and test text files it is creating but it doesnt have data about images. what has to do solve this query
I have used you repo for yolov4 custom functions, but i can't follow this tutorial for custom data set cause it is using darknet and it is not the same folders as yolov4 custom function repo, it would be great if you could point out where can i see this
Let me know which custom classes you are training your custom detector on! As always comment if you run into any issues. Cheers :)
how to count plate in real time video sir?
Would love a video on Deep Sort + YOLOv4
I started training dataset containing wheat images in colab and colab's runtime is just stopped at 1961 iteration inspite of setting setting javascript code in console. Can you please suggest any way we can prevent it so that training continues?
prateek gupta that can sometimes happen due to memory issues within the notebook after running for a long time. Just restart training from saved weights and you will be good as new!
When I am running the first step in step 5, it shows this. What should I do?
CUDA Error: out of memory
CUDA Error: out of memory: File exists
darknet: ./src/utils.c:326: error: Assertion `0' failed.
I just saw that this video was posted. I was literally following your tutorials over the last few days to build a license plate detector for my University Major Project. You may have just saved my life, I'll be watching this video later
Perfect timing haha. Hope it helps, let me know if you need any help!
@@TheAIGuy An update on my project. Everything went well in my presentation besides a few questions on logistics. Thanks to you, it looks like I've passed my Major Project, the most important module to graduate from my university in Jamaica. All that's left to do is finish 2 modules to meet the credit requirements to graduate. Thank you for your video, it really helped me out of a tough spot
@@brandonchung5305 hello! Would you be open to sharing the code? I'm working on a similar project and i am a bit stuck. Or if possible, would you be open for a chat??
@@CJ-wk1xj Hey, send me a message on Discord at DeltaHarbinger#7582. I'll respond in a few hours
@@brandonchung5305 i just did :)
You teach better than paid courses which I've been taking for a couple of months. Please keep up the good work, thank you very much, you're teaching me a lot!
This is exciting, I was finally able to use your tutorial to train and test a working detector! you are making machine learning accessible to someone without a background in it, and I appreciate that. I only trained a small model (I did a beer detector lol), but this will be a great foundation for a larger paper I'm working on for my undergraduate final project. Thanks again!
Hands down the most concise and practical YOLO tutorial I've seen anywhere!
You're such a life savior. Stay Blessed !
Waiting for this so far! Thank you!
This is such a useful video. Extremely helpful and well documented notebook:)
Glad it was helpful! Thanks for the kind words.
Thanks, this tutorial still works as of 2/28/2022. The only issue is that Step 5 may not work on a free GPU instance. Google Colab offers plans to access High-RAM GPUs at $9.99/month. I bought the subscription, changed the runtime to a High-RAM GPU, and Step 5 worked for me.
You are a life saviour! I have a senior project and had to train a model on custom data, your notebook and explanation literally saved me! Fr! I learned from you more than 3 college courses of ai, thank you so so much
Hi! First of all I want to thank you for all your videos and the way you share your knowledge, that is invaluable and appreciated!
A while ago, I trained a single class with YOLOv3 to detect Argentine license plates (the old ones and the new ones) with 340 images. Of the 340 images you can detect 339, there is always one that does not detect it, it does not matter. I have done it because I wanted to try a Raspberry Pi and a webcam, do license plate recognition to recognize the characters and save them in a database, and that's where I am limited with my knowledge. Once I recognize where the patent is and draw a frame around it (already done), I need to extract it, correct its perspective, and do the OCR. I have seen that there exists YOLO, Tiny YOLO, Tensor Flow, Kereas, etc, and all of them using OpenCV; and the truth is that I already got lost. In your experience, what would be the ideal path to follow for a modest project like the one I want to do? Thank you very much.
I would recommend OpenCV with python. You can save the YOLOv4 detections and then run the OCR on the saved image. I will try and create a video showing how to do this on my license plate detector.
@@TheAIGuy I'll be looking forward for your video!
@@TheAIGuy can you make relatime ocr? so yolov4 detection and make it to OCR too iin one network
@@TheAIGuy Please make video on extracting licence plate only from this image and applying OCR pytesseract on same..also for video as well.. please sir
@@TheAIGuy Sir, your videos are really helpful.i really appreciate your efforts and work.
Thank you so much
You just made my day even better ! =)
One of the most useful and comprehensive videos I have watched in a long time. A lot of videos leave out important detail. Thank you.
wow!! such value in every video! huge fan of your page: )
Thanks a lot. Been stuck at this for last few days. Great video. The backup part is awesome.
Best Content and Best Explanation on Internet, Immediately subscribed to your channel when u started explaining the datasets making ..... BEST
amazing tutorial dude! You smashed DEEPLEARNING into a piece of cake!
Thank you! your tutorial is definitely one of the best if not the best!.
Thanks so much! Much appreciated.
Man you are just awesome...couldn't find multiple subscribe button..Thank you so much
I would like to thank you for this video and the collab notebook you single handedly saved my final year project. And i hope you are doing well man. Keep up the good work you are the best
Thanks bro .Your videos are just amazing .I just want to suggest that kindly make video regarding test data on custom trained model and also a video regarding description of the output of yolo v4.Other than that you are just amazing
Appreciate it. Thanks for the great ideas, noted them down for future videos.
@@TheAIGuy So nice of you. Great work my friend👌🏻
Thanx man, it works like a charm. Looking forward for more tutorials
Glad you got it working. Thanks for the support!
I really like the intro box animations; it's super cool
Hi !
Your tutorial is awesome. You have done a lot of hardwork to make this tutorial.
Thank you so much.
That was exactly what I was looking for!
thx mate :)
This video has helped me a lot :D. Congratulations on the tutorial!! Will you make a video on how to build an Object Tracker using YOLOv4? I think it would be interesting and would help a lot of people to develop different applications.
Great idea on the object tracker with YOLOv4! Will definitely cover this in a future video!
I second this!
Man your notebook was very very helpful to me cheers...
Man, I love you! You have no idea of how much you help me! Thanks!
Here we are! My Guy* with another trendy trend huh? Good job, thanks for the share! :D
I got follwing error while I ran !python \convert_annotations.py command
error is
Currently in subdirectory: validation
Converting annotations for class: Vehicle registration plate
0% 0/30 [00:00
Hey did you get your solution??
New content I got from your video, thanks 4 sharing.
Thank you! your tutorial is definitely one of the best if not the best!. helped me a lot
Thanks!! Great TH-cam videos and you are very good at explaining
I am having an issue, the generate_test.py and generate_train.py scripts are generating empty train and test.txt for me.
My friend you are a very good person.
Thank you so much for making such a great video ! I learned a lot in this video, I gain lots of practical experience from it. Thank you so much !!!!
I need to thank you for your job, it's obvious that this didn't worked everything as expected and I ended up solving many problems all along the training, but that's how we make science, discovering and understanding; Of course that without you it would be a LOT harder, thank you, I finally trained my first model!
It was a great tutorial.
Thank you so much...!
I dont understand the cd darknet part on 6:14
I dont have this folder on my drive. What should I do about this?
As usual a really helpful tutorial, just wanted a suggestion how to store all the detected object co-ordinates from the video in a Jason file for further processing.
Love your tutorials much better then the paid udemy tutorials
Really thank you!! It have been help me a lot in my final paper!
You deserve a million subscribers.
I love your tutorials. they are just amazing. If possible could you please show training of GPT , BERT or XLnet
Thanks for the very clear explanation.
wow thanks for making such a great tutorial.
I try to do it with my custom dataset with 38 classes and always get "CUDA status Error: file: ./src/dark_cuda.c : () : line: 373 : build time: Sep 22 2020 - 09:47:07
CUDA Error: out of memory
CUDA Error: out of memory: File exists
darknet: ./src/utils.c:325: error: Assertion `0' failed.", what can be the reason? I've already set up division for 64 and batch for 2 images and random to 0 etc.
@AI Guy , training stopped after create 6 cpu-threads. Can you help me ?
Thank's for a video, I want ask something , how to change conf_threshold and IoU_Threshold went I want to check my model mAP (step 6)?
Amazing video, very clear ! Thank you
Fantastic you nearly done my entire final year project😁, would you do YOLOv4 and Deep SORT for multi-object tracking in the future? I know you did YOLOv3 and Deep SORT and you nailed it.
Will definitely cover this soon! Thanks for the support.
I did training on yolov4 on google colabs. and I only get last weight no backup files per 1000 iterations. How can be like that?
fyi, I did training for 36 classes, and I don't think there's anything wrong with my backup file settings. is it okay? or i made mistake?
thank u before!
thank you, I got to learn so much from this video!!!
Thanks a lot for this amazing tutorial, I really appreciate your efforts in preparing this wonderful video and this detailed notebook, god bless you :)
I have followed all the steps and mounted the drive properly hence gdrive appears on the file manager tab. However, while training I have followed the mydrive/yolov4/backup in both obj.data and created the folders in the drive but the trained weights are not saved. If I load a weights file/pre trained weights it can load it properly and resume training but it does not save the weights after 100 iterations.
46:32 cam you try it with webcam slanted photo taken in shade on sunny day while car is moving ?
What changes are required if I want to train yolov3-tiny and yolov4 tiny?
You would want to edit yolov4-tiny.cfg inside the cfg/folder for the custom cfg. As well you will need to download the initial tiny weights for training instead of the .conv file used for full yolov4.
@@TheAIGuy can you provide the links for both yolov3-tiny and yolov4-tiny initial weights link from where I can download these weights. Thanks
This video has saved my life! Much respect, thanks a lot for all the time and effort you put into your videos.
Ive trained a model to identify a pallet of boxes. I have a question, would it be possible to have a video and track an object, in my case a pallet, and then perform a some edge detection or detect another class of objects ONLY within the bounding box of the identified pallet? Basically I'm wanting to count the number of boxes on the pallet and I have it working with images, but video is the next step.
Any advice would be awesome.
Have you tried YOLO+ deepsort?
THis is really helpful, thank you so much.
@The AI Guy Thanks a lot for this video. I have trained the model on Custom Dataset of Retail Receipt Detection. It's working fine.
But I think my model is over fitted. I have early stop the training at 1100 iteration.
This is the MAP of my checkpoint.
Done! Loaded 162 layers from weights-file
calculation mAP (mean average precision)...
Detection layer: 139 - type = 27
Detection layer: 150 - type = 27
Detection layer: 161 - type = 27
152
detections_count = 172, unique_truth_count = 150
class_id = 0, name = retailReceipt, ap = 99.09% (TP = 149, FP = 2)
for conf_thresh = 0.25, precision = 0.99, recall = 0.99, F1-score = 0.99
for conf_thresh = 0.25, TP = 149, FP = 2, FN = 1, average IoU = 83.67 %
IoU threshold = 50 %, used Area-Under-Curve for each unique Recall
mean average precision (mAP@0.50) = 0.990915, or 99.09 %
Total Detection Time: 8 Seconds
I have a question. I already follow your tutorial on building object tracking using yolov3 and it works. But how do I get some kind of result on the tracking in terms of value or graphs? I need to evaluate the tracker output. I hope you can help me with that. Thank you
Hi Bro, a cheers from HK! I hv followed every steps in yr tutorial, using a trained weight model (yolov3_8000) from others instead of yr pre-trained weights.
When I directly do Step 7: Run Custom Object Detector without train it on Colab, it shows "Unable to init server: Could not connect: Connection refused "
When I go to check the mAP of his model, it shows "Allocate additional workspace_size = 52.43 MB,
Loading weights from /.weights...Couldn't open file: /.weights ".
Can u advise how to solve this? Really really appreciate to u, bro!!!!
Very useful video!
Thank you!
Thanks for the video, I learned alot from it. Just wondering how can I change so the prediction image doesn't only include border but it also shows the percentage of confidence.
Let’s say yolo is detecting multiple objects of the same class, is there a way to conditionally stop detecting one of those objects?
yes check the *pyimagesearch* blog in which opencv has been used to read from the trained yolo model, subsequently taking the raw output and accepting only the detections related to some specific class.
Amazing vdo sir ❤️
After I start the training everything goes fine but when it prints out
Detection layer: 161 - type = 28
Resizing, random_coef = 1.40
608 x 608
Create 6 permanent cpu-threads
It stops. Please help.
Thanks.
Same problem here tell me once you get solved
Check the train.txt and test.txt files (darknet > data > train.txt) if it's empty then something went wrong with the python scripts. I had the same problem because I manually labeled the images and used png images instead of jpg. So I had to replace jpg with png in the generate_train.py
I am training a detector for 'Persons'.
My detection and validation datasets are 5000 and 1500.
How many iterations will it take to bring the average loss down to 0.5 or mAP up to 95% ?
Hi am trying to train on desktop when am started training am getting the error message like could not load images even the paths are correct in train.txt any thoughts what maybe the issue ?
Sir getting an error during training after building conv layer error is Could not open file :'data/names,'
Hi, I could learn so many things in YOLO through your TH-cam!
Now, I also want to see the tutorial on applying the model to the webcam.
Do you have a plan to make tutorial about applying the model to the webcam?
I hope so!
What if we wanted to just add 1 more class to the ones YOLOv4 can already detect? In particular I would like to detect both "people" and "faces" using the custom functions you made in a later video for detecting and cropping these 2 classes.
- What changes would I need to make to classes.txt, obj.data and obj.names?
- And to the variables __C.YOLO.CLASSES (coco.names) and __C.TRAIN.ANNOT_PATH (val2017.txt) of the config.py file of your custom functions?
I lack some experience in this field and I'm not sure whether I should retrain the whole model to detect these 2 classes from scratch or if I can just retrain the model so that it can learn to detect faces while still being able to detect people as it can by default. Love your content, the most helpful and practical I've seen :)
Great tutorial! thanks!
would really like to see you do this tutorial for a local machine using CPU or GPU options. Having trouble implementing on my local pc. Thanks.
Hi AI guy can we measure the area (surface) of an object in video with Yolo or any other program
Hi The AI Guy,
Great video! I have a couple of questions.
1. Is there a way I can save the chart.png that shows both avg loss and map, to my google drive?
2. My weights saved every 10k iterations instead of 1k iterations, where would I make this change to make sure that the weights are saved every 1k iterations instead of 10k iterations?
3. Any recommendations/suggestions to keep in mind for OCR using YOLOv4?
Thank you so much man🥺 god bless you
Do we have to do any preprocessing to the images dataset augmentation before performing the steps or should we just start annotating the images and follow the steps from the video
Do a playlist for Yolov5 as well please...
Hello, I just have a question
Is GPU Nvidia gt 840m compatible for yolov4 ?????
another great video!!thanks!!!
Thanks for the video. When trying to unzip the files I get this error:
Archive: ../test.zip
warning [../test.zip]: 496177718 extra bytes at beginning or within zipfile
(attempting to process anyway)
error [../test.zip]: start of central directory not found;
zipfile corrupt.
(please check that you have transferred or created the zipfile in the
appropriate BINARY mode and that you have compiled UnZip properly)
(and same for obj.zip)
Any ideas on what went wrong? I already made a few models following your tutorial for yolov3 in Cloud and never ran into this error.
I'm trying to train just one object but I keep getting error, "Wrong annotation: class_id = 15. But class_id should be [from 0 to 0]".
Any thoughts?
I'm getting similar error: Wrong annotation: class_id = 101. My .txt all start with 101. Seems like more than just a coincidence. Not sure what I'm doing wrong or how to fix. Any idea Mr AI Guy? This video is thoroughly awesome.
@@brentdunklau4536 I think I solved this as my model seems to be training OK.
Before you start labelling find the directory where LabelImg is installed on your computer. Inside that folder is a predefined_classes.txt. Delete all the entries inside this .txt file.
What I think is happening is LabelImg comes with pre defined classes like dog, person, cat ect. There are 16 in all hence my wrong annotation class_id of 15 (LabelImg starts at 0). You have probably got 85 objects in your labelling but the model is looking for 101 including the dog, person, cat ect. I'm sure there's a way you can get around this without having to re-labelling all your data.
Hope this works for you.
Edit, this solved my problem.
How do i delpoy this in my project?...can i use REST api to get the output from colab?
One of the best videos on yolov4...
Thanks for explaining it so well
I have a small doubt... can we change the batch_size=2 and image_size=1024??
If we can do it what should be the value of subdivision??
When you retrain YOLOv4 this way, does it use any transfer learning? In other words, do the previous weights carry over to the new training and the new training does not retrain the model from scratch?
I get this error when i try to train with one class. I think there are places where the classes=1 to be set also some other changes. Can you share the .cfg file for training the above example with one class.
Error: in the file data/obj.names number of names 1 that isn't equal to classes=80 in the file cfg/yolov4-obj.cfg
open your .cfg file and search for classes. You should find it there like 3 times, the default setting is 80, I would assume you didn't changed it
@@mattishamann8660 i am also facing that issue,while i have changed 3 classes which was =80 ,than =1 as per recommendation,how to get solve please inform
on training block execution ....
Couldn't open file: data/obj.data
what should i do ????
I'm also having a similar problem with the data/train.txt file. Has anyone found a solution?
I wanted to add I was able to varify the train.txt file was in the darknet data folder on my VM, google colab. The first line reads >> data/obj/0145644b6378f2ea.jpg
No
what about the test data ? you did not use them even in last step (step7),shall we use them in this step?
Lets help each other out and share custom detectors. Message each other if someone has something you need. Comment what you are training here:
Turtles
Overhead people images
You're a god, thank you so much!
Hello, I found this article to be excellent. May I inquire about the limitations regarding custom datasets? Specifically, I have prepared 65,000 labeled images along with corresponding text files. Would this dataset size be acceptable on google colab?
Awesome tutorial! Class!
btw, how to show the probability of each bounding box?
I have created my custom dataset with the YOLOv4 compatible annotations. I have a question to ask.
1. Where does the classes.txt file has to be placed?
I was trying to train yolov5 on custom data but when giving command for training , it is just exiting the cell after below message
"Saving weights to backup/yolo-obj_final.weights
Create 6 permanent cpu-threads
If you want to train from the beginning, then use flag in the end of training command: -clear "
I am using multi class dataset in the generation of train and test text files it is creating but it doesnt have data about images. what has to do solve this query
when I reopen google collab on next day , why I have to re-clone darknet every time? I saved it but it doesn't work
Great Video! So in the future after I created the initial model, how would I be able to add more classes to the model?
I have used you repo for yolov4 custom functions, but i can't follow this tutorial for custom data set cause it is using darknet and it is not the same folders as yolov4 custom function repo, it would be great if you could point out where can i see this
Am i able to change anything within the final line where youre testing your custom detector on images to videos instead?