i am having issue (yolov3-gpu) D:\Object-Detection-API-master>python load_weights.py 2020-04-10 12:54:26.289940: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudart64_101.dll Traceback (most recent call last): File "load_weights.py", line 4, in from yolov3_tf2.models import YoloV3, YoloV3Tiny File "D:\Object-Detection-API-master\yolov3_tf2\models.py", line 23, in from .utils import broadcast_iou File "D:\Object-Detection-API-master\yolov3_tf2\utils.py", line 5, in from seaborn import color_palette File "C:\Users\ibrah\anaconda3\envs\yolov3-gpu\lib\site-packages\seaborn\__init__.py", line 6, in from .rcmod import * File "C:\Users\ibrah\anaconda3\envs\yolov3-gpu\lib\site-packages\seaborn cmod.py", line 6, in from . import palettes, _orig_rc_params File "C:\Users\ibrah\anaconda3\envs\yolov3-gpu\lib\site-packages\seaborn\palettes.py", line 10, in from .utils import desaturate, set_hls_values, get_color_cycle File "C:\Users\ibrah\anaconda3\envs\yolov3-gpu\lib\site-packages\seaborn\utils.py", line 7, in from scipy import stats File "C:\Users\ibrah\anaconda3\envs\yolov3-gpu\lib\site-packages\scipy\stats\__init__.py", line 384, in from .stats import * File "C:\Users\ibrah\anaconda3\envs\yolov3-gpu\lib\site-packages\scipy\stats\stats.py", line 179, in from scipy.spatial.distance import cdist File "C:\Users\ibrah\anaconda3\envs\yolov3-gpu\lib\site-packages\scipy\spatial\__init__.py", line 99, in from .qhull import * ImportError: DLL load failed: The file cannot be accessed by the system. why 😫, and how to solve it,
This is really amazing work. Thank you so much. Could you send the tracker information to a note pad so that I can take that information and use it after effects for example?
Object tracking looks awesome! Thanks for the yolov3 videos, they are life-saviors! Love to see a video about training custom model directly on TF 2.0 instead of converting one from darknet. Much appreciated!
Hey before anything!!!! Thanks for your videos! I have learned a LOT. I really appreciated it. The object tracking is an amazing feature and seeing this evolving to memorizing an object based on a feature would be super cool! Best Francisco
great tuitorial ,after 3 weeks of trying finally something worked on my system,thanks a lot , and would love to see a vedio in which it can generate any kind of alarm after detection from web cam
Fix for ValueError: cannot reshape array of size 281082 into shape (256,128,3,3) Just chack your number of classed in obj.names and yolov3-custom if they both match that error should not come and make sure you point out to the yolov3-custom file
Thank you for the great video, it helped me alot in optimizing the live object detection. Before the video I always used my CPU and not my GPU but now it's not 1 fps anymore ^^ Thank You
I'd like to post the solution to my prob as The AI Guy replied to some one else: " try running the flags without quotes around them like and maybe add a ./with the first folder -video ./data/video/video.mp4, try that out"
Hi The AI Guy, thanks for the video! But I have got a problem while loading my custom weights. When I added my obj.names file into the labels folder and converted my weights file into tensorflow file, the detection is working, but the video is detecting from the coco.names classes. So, I moved coco.names to another location and kept only my obj.names file in the labels folder, and I got the following error after I ran the detection. Please help ! python detect_video.py --weights 'weights/yolov3-custom.tf' --output 'data/video/maskoutput.avi' --video 'data/video/mask-video.avi' -num_classes '2' data/labels/obj.names 2020-06-15 20:43:43.381309: I tensorflow/core/platform/cpu_feature_guard.cc:142] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 I0615 20:43:51.600792 14012 detect_video.py:36] weights loaded Traceback (most recent call last): File "detect_video.py", line 94, in app.run(main) File "C:\Users\lenovo\Anaconda3\envs\yolov3-cpu\lib\site-packages\absl\app.py", line 299, in run _run_main(main, args) File "C:\Users\lenovo\Anaconda3\envs\yolov3-cpu\lib\site-packages\absl\app.py", line 250, in _run_main sys.exit(main(argv)) File "detect_video.py", line 38, in main class_names = [c.strip() for c in open(FLAGS.classes).readlines()] FileNotFoundError: [Errno 2] No such file or directory: './data/labels/coco.names' (yolov3-cpu)
HI AI Guy, After converting weights into TF models, I have ensured that all required packages are been installed by running: pip install -r requirements.txt I tried running the model to detect video by the code which you have provided in MY 'CPU'- python detect_video.py --video 'data/video/video.mp4' --output 'data/video/output.avi' Upon running the code, i got the below error : [ERROR:0] global C:\projects\opencv-python\opencv\modules\videoio\src\cap.cpp (392) cv::VideoWriter::open VIDEOIO(CV_IMAGES): raised OpenCV exception: OpenCV(4.1.1) C:\projects\opencv-python\opencv\modules\videoio\src\cap_images.cpp:253: error: (-5:Bad argument) CAP_IMAGES: can't find starting number (in the name of file): 'data/video/output.avi' in function 'cv::icvExtractPattern' W0325 20:00:54.601627 14240 detect_video.py:63] Empty Frame W0325 20:00:54.704001 14240 detect_video.py:63] Empty Frame W0325 20:00:54.806705 14240 detect_video.py:63] Empty Frame Please help.
@@TheAIGuy It worked AI Guy... However, the model detects at 0.5 fps!!!... any remedies for it? else is it because i am using CPU, the fps goes down to this range?
How to connect darknet with python... Actually I am doing a project .. which is basically identifying the censorable contents like smoking detection, driver wearning seatbelt , motorcyclist wear helmet and alcohol detection in movies.. and automatically show the statutory warning on the corresponding frames... I use yolov3 for detection.. and used darknet framework... I need to add an overlay image in corresponding frames..how to implement this..
*Sir your videos are super helpful 😁, could you provide a code to extract a particular class image after yolov3 detects it, suppose I have custom dataset yolo model that detects a particular animal ,can I just extract it's pic every time yolo detects it!*
okay I will add this to my next video! Thanks everyone for feedback. I will add that you can select classes and take a screen capture when these classes(s) are detected.
Man you are AMAZING! You just saved me and my project, because I was started to giving up but when I see objects are recognized on the webcam I took a deep breath, really thank you very much! I just wonder if I can see the output of the detected objects, ie: if camera detects a cellphone at x=20,y=50, it should print like: "at blabla frame cellphone at x=20,y=50" Is something like this achivable? if yes where should i edit the code, thanks again. You just earned a new subscriber
Yes this do-able for sure1 You will want to edit detect_video.py after line 76. Line 76 is where you get all the box coordinates, confidence values and class names. You can easily write a print statement to print what you are looking for with this information!
hi great stuff man helped a lot! tho would love to see tutorials on code explanations like internal workings of yolo (in terms of code) some explanations of files in the folder and detections
Hey great video! I had an issue with doing the initial detecting. The error that came up was this 2020-04-22 12:49:41.025973: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'nvcuda.dll'; dlerror: nvcuda.dll not found 2020-04-22 12:49:41.033458: E tensorflow/stream_executor/cuda/cuda_driver.cc:351] failed call to cuInit: UNKNOWN ERROR (303) not quite sure how to fix it :\
Yes there is. You can edit the code in detect_video.py in order to perform this. look for the variable "boxes" as they are the bounding box coordinates. Then just opencv to crop!
Hello, I really like your in depth tutorials, they are very helpful. I successfully followed your tutorial about training YOLOv3 custom model on Cloud, I downloaded the .weights file and convert it to .tf model following this tutorial but when I use it for on-video object detection I get this error: raise ValueError("Shapes %s and %s are incompatible" % (self, other)) ValueError: Shapes (255,) and (42,) are incompatible any ideas how to solve this issue ?
I think this has to do with the number of filters edited in the cfg file ( before training i changed the default value of filters (which is 255) to 42 as i am training the model on 9 classes of objects)
Hi saw your video and liked it a lot. I have a doubt: How can I add the path of a video that I am getting in real time from ROS - its the real time camera feed of a drone in gazebo. How do I add its path?
great tutorial, just wondering, if i have to add multiple cameras inside the detection_video.py, what should i change in order for it to work? i know i can add cv2.Videocapture(0) and cv2.Videocapture(1). but im not too sure about everything else.
thanks for this very clear video :) but i have 2 questions. 1.)how can we adjust position of output frame, half of it out of screen. and 2.) i have this error : warning: Error opening file (/build/opencv/modules/videoio/src/cap_ffmpeg_impl.hpp:856) and also i can't write output video please help me...
i am sorry i recognized that i don't have opencv-python. but after i did ' pip install', i have this error : [ERROR:0] global C:\projects\opencv-python\opencv\modules\videoio\src\cap.cpp (392) cv::VideoWriter::open VIDEOIO(CV_IMAGES): raised OpenCV exception: OpenCV(4.1.1) C:\projects\opencv-python\opencv\modules\videoio\src\cap_images.cpp:253: error: (-5:Bad argument) CAP_IMAGES: can't find starting number (in the name of file): 'C:/Users/User/Desktop/AIGUY_videos/Object_Detection_API_master/data/video/output.avi' in function 'cv::icvExtractPattern' can you help me??
Sir I have tried out a one line code similar to that could detect object on images ,on a video and saved it in similar way you did now but the problem was that I wasn't able to use that code for my custom yolo v3 detector, is there something I need to do? I have made the custom yolo detector using your previous video❤️ in colab ,can you suggest any solution?
Hi, Thank you very much for an awesome video. Liked and Subscribed. There is a question I would like to ask. I saw in the label folder there is a coco text file with all the labels of the coco dataset. I try to reduce it to only certain items but it doesn't work. Could you suggest to me a way to do so?
Hi when my trying to convert my custom weights(with classes 2) to tensorflow m getting this following error, can you please tell me the solution File "C:\Users\hp\Desktop\Object-Detection-API-master\yolov3_tf2\utils.py", line 68, in load_darknet_weights conv_shape).transpose([2, 3, 1, 0]) ValueError: cannot reshape array of size 75571 into shape (256,512,1,1)
TAG, Another great video! I'm a big fan! Questions on the '--video' webcam flag, if the '0' is replaced with and IP address ( to an IP camera) would the code still work? Thanks and looking forward to your next video.
Hey! I'm appreciate for your another great video. and Im having an error in the process of loading my custom weight file. Error message it "ValueError: cannot reshape array of size 73122 into shape (256,128,3,3) (yolov3-cpu)". I have set the command like python load_weights.py --weights './weights/yolov3_custom_last.weights' --output './weights/yolov3-custom.tf' --num_classes 1. But getting this error. How should I fix this?
Try this command: python load_weights.py --weights ./weights/yolov3_custom_last.weights --output ./weights/yolov3-custom.tf --num_classes 1 --tiny; Also becase you did custom weight file, you need search num_classes in visual studio code and replace all num_classes = 1.
Is it possible to remove classes from the already pre-trained yolov3 model?? For example i want to only detect persons but not the other classes without retraining the model, is this possible?
It is not possible to remove classes from the train model, however you could add logic in the code to only show the detections for persons. This is the easiest way. You would just create an if statement and have it make sure the class is person before drawing detections.
tensorflow/core/platform/cpu_feature_guard.cc:142] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 I'm getting this error
Why when I activate the CPU or GPU code The program give me error : DLL load failed : the file cannot be accessed by the system But when I do not activate any of those The program run just fine with the GPU ????
Hi, how do we get labels while we are detecting with camera, we are trying to build this app for blind people, it will detect objects and tell what it is so im trying to use your model and stuff now i need to get labels while real time detecting it. any help appriate it . respect
will it work with CPU 2.2GHz and 8GB RAM this real-time webcam detection? I am planning on deploying ur code in a small shop. Also does it only work on ubuntu? can we run python executable from ubuntu in windows?
I have problem running load_weights.py with my custom weights. When I set the number classes which is 1 in my case I get the error: FATAL Flags parsing error: flag --num_classes='1': invalid literal for int() with base 10: "'1'". I tried to use another number I still get the error. Thank you in advance. Great videos!
off the top of my head though you probably set --num_classes '1', and for some reason its reading it as a string so try putting no quotes --num_classes 1
@@TheAIGuy You were right by not including the ' ' in the command also for the file path it worked. I'm using the Anaconda prompt not sure if that could be the reason. Thanks for the responds!
First of all i wanna thank you so much for your great work and effort I really appreciate what you do❤️ I have a question, is this code work on google colab in the same way ?? And thank you❤️
when im trying to run this line "!./darknet detector train /content/darknet/data/obj.data /content/darknet/cfg/yolov3-custom2.cfg darknet53.conv.74 -dont_show" in colab it shows like this "/bin/bash: ./darknet: Is a directory". please help me
this is because you must be in the wrong directory. The command only works in the darknet directory so just run "%cd darknet" in the line above your training command and it should now work!
Hi AI guy, After converting yolov3_custom weights to tensorflow model, by running the below code - python detect_video.py --weights ./weights/--yolov3-custom.tf-- -- video 0 -num_classes 2 --classes ./data/labels/obj.names It shown me the below error at last: raise ValueError("Shapes %s and %s are incompatible" % (self, other)) ValueError: Shapes (255,) and (21,) are incompatible Please help.
Anyone know how to fix this error: Shapes error. File "C:\Users ob26\anaconda3\envs\yolov3-gpu\lib\site-packages\tensorflow_core\python\framework\tensor_shape.py", line 1115, in assert_is_compatible_with raise ValueError("Shapes %s and %s are incompatible" % (self, other)) ValueError: Shapes (255,) and (18,) are incompatible
So I have successfully trained my own dataset using the training in the cloud video and have obtained my pretrained weights file. I ran the following command python load_weights.py --weights ./weights/yolov3_custom_last.weights --output ./weights/yolov3-custom.tf -num_classes 9 and recieved the folowing error: Traceback (most recent call last): File "load_weights.py", line 34, in app.run(main) File "C:\Users\tronn\Anaconda3\envs\yolov3-gpu\lib\site-packages\absl\app.py", line 299, in run _run_main(main, args) File "C:\Users\tronn\Anaconda3\envs\yolov3-gpu\lib\site-packages\absl\app.py", line 250, in _run_main sys.exit(main(argv)) File "load_weights.py", line 21, in main load_darknet_weights(yolo, FLAGS.weights, FLAGS.tiny) File "C:\Code\Object-Detection-API\yolov3_tf2\utils.py", line 68, in load_darknet_weights conv_shape).transpose([2, 3, 1, 0]) ValueError: cannot reshape array of size 49334 into shape (256,128,3,3)
The AI Guy thank you. I didn’t let the whole training Finnish in the cloud. Stopped it at about 2k something iterations, I did test it on the cloud model and it did work. Used those weights on the conda environment and got that error. Also- I only used 49 images as my original dataset amount. Another thing - could there be some sort of mismatch with the config file and the python script that creates the tensor?
SO I gave in and went the Linux route. I was up until 3:00am in the morning but I have darknet and tensorflow all working perfectly. I am going to try to retrain my custom dataset in this environment now and will let you know how it went. Also, my problem might have been due toi an extra space in my classes.txt file. Going to make sure that is not the case.
Hi everyone! Let me know if you have issues here in the comments as well as what is your favourite class to detect!
I don't see your object_tracker.py in your repo. Am I missing something?
@@tomboa Its not finished yet, will be out soon!
The AI Guy Why does the video work so slowly in the detection process?
i am having issue
(yolov3-gpu) D:\Object-Detection-API-master>python load_weights.py
2020-04-10 12:54:26.289940: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudart64_101.dll
Traceback (most recent call last):
File "load_weights.py", line 4, in
from yolov3_tf2.models import YoloV3, YoloV3Tiny
File "D:\Object-Detection-API-master\yolov3_tf2\models.py", line 23, in
from .utils import broadcast_iou
File "D:\Object-Detection-API-master\yolov3_tf2\utils.py", line 5, in
from seaborn import color_palette
File "C:\Users\ibrah\anaconda3\envs\yolov3-gpu\lib\site-packages\seaborn\__init__.py", line 6, in
from .rcmod import *
File "C:\Users\ibrah\anaconda3\envs\yolov3-gpu\lib\site-packages\seaborn
cmod.py", line 6, in
from . import palettes, _orig_rc_params
File "C:\Users\ibrah\anaconda3\envs\yolov3-gpu\lib\site-packages\seaborn\palettes.py", line 10, in
from .utils import desaturate, set_hls_values, get_color_cycle
File "C:\Users\ibrah\anaconda3\envs\yolov3-gpu\lib\site-packages\seaborn\utils.py", line 7, in
from scipy import stats
File "C:\Users\ibrah\anaconda3\envs\yolov3-gpu\lib\site-packages\scipy\stats\__init__.py", line 384, in
from .stats import *
File "C:\Users\ibrah\anaconda3\envs\yolov3-gpu\lib\site-packages\scipy\stats\stats.py", line 179, in
from scipy.spatial.distance import cdist
File "C:\Users\ibrah\anaconda3\envs\yolov3-gpu\lib\site-packages\scipy\spatial\__init__.py", line 99, in
from .qhull import *
ImportError: DLL load failed: The file cannot be accessed by the system.
why 😫, and how to solve it,
This is really amazing work. Thank you so much. Could you send the tracker information to a note pad so that I can take that information and use it after effects for example?
hey. even thought this video is a year old it's still one of the best resources I've found. great work and thank you so much
Object tracking looks awesome! Thanks for the yolov3 videos, they are life-saviors! Love to see a video about training custom model directly on TF 2.0 instead of converting one from darknet. Much appreciated!
Appreciate it! I will try and get this done!
i just wake up and booom another good video come out ahhh thank bro you dont know how i feel learning from you my best youtube chanel
Appreciate it my man! Thanks a lot.
Hey before anything!!!! Thanks for your videos! I have learned a LOT. I really appreciated it. The object tracking is an amazing feature and seeing this evolving to memorizing an object based on a feature would be super cool! Best Francisco
Appreciate it! The object tracking video should be out pretty soon. Stay tuned!
Great video for beginners, clear explanation. Thank you
Wow man, thanks!! youre work really made it a whole lot easier to work with object detection and yolo
great tuitorial ,after 3 weeks of trying finally something worked on my system,thanks a lot ,
and would love to see a vedio in which it can generate any kind of alarm after detection from web cam
Thanks. That is a cool idea!
Fix for ValueError: cannot reshape array of size 281082 into shape (256,128,3,3)
Just chack your number of classed in obj.names and yolov3-custom if they both match that error should not come and make sure you point out to the yolov3-custom file
Muchas gracias ! El tutorial más empático, paso a paso y que funciona que he visto de yolo
Thank you for the great video, it helped me alot in optimizing the live object detection. Before the video I always used my CPU and not my GPU but now it's not 1 fps anymore ^^ Thank You
This tutorial is super helpful, thanks!
Any reason why it works with pics, but when i try the video it says "detect_video.py:63] Empty Frame" and gives an empty video as output?
I'd like to post the solution to my prob as The AI Guy replied to some one else: " try running the flags without quotes around them like and maybe add a ./with the first folder -video ./data/video/video.mp4, try that out"
Hi Thanks to you, I learned how to use yolov3. cheers mate :)
Hi The AI Guy, thanks for the video! But I have got a problem while loading my custom weights. When I added my obj.names file into the labels folder and converted my weights file into tensorflow file, the detection is working, but the video is detecting from the coco.names classes. So, I moved coco.names to another location and kept only my obj.names file in the labels folder, and I got the following error after I ran the detection. Please help !
python detect_video.py --weights 'weights/yolov3-custom.tf' --output 'data/video/maskoutput.avi' --video 'data/video/mask-video.avi' -num_classes '2' data/labels/obj.names
2020-06-15 20:43:43.381309: I tensorflow/core/platform/cpu_feature_guard.cc:142] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2
I0615 20:43:51.600792 14012 detect_video.py:36] weights loaded
Traceback (most recent call last):
File "detect_video.py", line 94, in
app.run(main)
File "C:\Users\lenovo\Anaconda3\envs\yolov3-cpu\lib\site-packages\absl\app.py", line 299, in run
_run_main(main, args)
File "C:\Users\lenovo\Anaconda3\envs\yolov3-cpu\lib\site-packages\absl\app.py", line 250, in _run_main
sys.exit(main(argv))
File "detect_video.py", line 38, in main
class_names = [c.strip() for c in open(FLAGS.classes).readlines()]
FileNotFoundError: [Errno 2] No such file or directory: './data/labels/coco.names'
(yolov3-cpu)
Great Video. Really a nice explanation and I got a very clear picture of whole scenario. Keep it up! 5 stars!..
HI AI Guy,
After converting weights into TF models,
I have ensured that all required packages are been installed by running:
pip install -r requirements.txt
I tried running the model to detect video by the code which you have provided in MY 'CPU'-
python detect_video.py --video 'data/video/video.mp4' --output 'data/video/output.avi'
Upon running the code, i got the below error :
[ERROR:0] global C:\projects\opencv-python\opencv\modules\videoio\src\cap.cpp (392) cv::VideoWriter::open VIDEOIO(CV_IMAGES): raised OpenCV exception:
OpenCV(4.1.1) C:\projects\opencv-python\opencv\modules\videoio\src\cap_images.cpp:253: error: (-5:Bad argument) CAP_IMAGES: can't find starting number (in the name of file): 'data/video/output.avi' in function 'cv::icvExtractPattern'
W0325 20:00:54.601627 14240 detect_video.py:63] Empty Frame
W0325 20:00:54.704001 14240 detect_video.py:63] Empty Frame
W0325 20:00:54.806705 14240 detect_video.py:63] Empty Frame
Please help.
Akhil G try running the flags without quotes around them like and maybe add a ./with the first folder -video ./data/video/video.mp4, try that out
@@TheAIGuy You are awesome man!!! it worked!!!
@@TheAIGuy It worked AI Guy...
However, the model detects at 0.5 fps!!!...
any remedies for it? else is it because i am using CPU, the fps goes down to this range?
@@akhil4441 Yes due to using CPU the detection on video can be very slow. If you have a GPU i recommend running the model with GPU.
awesome video tutorial, one of the best, thanks for sharing !!
Thanks for the kind words, glad you enjoyed!
How to connect darknet with python...
Actually I am doing a project .. which is basically identifying the censorable contents like smoking detection, driver wearning seatbelt , motorcyclist wear helmet and alcohol detection in movies.. and automatically show the statutory warning on the corresponding frames... I use yolov3 for detection.. and used darknet framework... I need to add an overlay image in corresponding frames..how to implement this..
I too need it
so when it detects certain classes it outputs text on screen?
@@TheAIGuy I need to add small statutory image on certain screen..
Akhil G Krishnan okay i will try and have this in my next video
*Sir your videos are super helpful 😁, could you provide a code to extract a particular class image after yolov3 detects it, suppose I have custom dataset yolo model that detects a particular animal ,can I just extract it's pic every time yolo detects it!*
Ya dude , I too needs it!😩
Same request here
Need the samething
okay I will add this to my next video! Thanks everyone for feedback. I will add that you can select classes and take a screen capture when these classes(s) are detected.
@@TheAIGuy I need the bounding box from video ,I need to process what's inside it
Please help 😊
Man you are AMAZING! You just saved me and my project, because I was started to giving up but when I see objects are recognized on the webcam I took a deep breath, really thank you very much! I just wonder if I can see the output of the detected objects, ie: if camera detects a cellphone at x=20,y=50, it should print like: "at blabla frame cellphone at x=20,y=50" Is something like this achivable? if yes where should i edit the code, thanks again. You just earned a new subscriber
Yes this do-able for sure1 You will want to edit detect_video.py after line 76. Line 76 is where you get all the box coordinates, confidence values and class names. You can easily write a print statement to print what you are looking for with this information!
@@TheAIGuy Thx! I will definitely try it
@@TheAIGuy How exactly would you change the code, so if object "Dog' detected, then do the following code? Thanks
Another amazing guide.
Hey, please do a tutorial on pose detection using PoseNet and comparison of posses.
I have been meaning to do this. I'll start working on it.
hi great stuff man helped a lot! tho would love to see tutorials on code explanations like internal workings of yolo (in terms of code) some explanations of files in the folder and detections
Hey great video! I had an issue with doing the initial detecting. The error that came up was this
2020-04-22 12:49:41.025973: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'nvcuda.dll'; dlerror: nvcuda.dll not found
2020-04-22 12:49:41.033458: E tensorflow/stream_executor/cuda/cuda_driver.cc:351] failed call to cuInit: UNKNOWN ERROR (303)
not quite sure how to fix it :\
why i can not install absl? it is not included in requirements.txt
you can install absl, i think it gets pre-downloaded as a requirement for one of the other dependencies.
The AI Guy whatever I use conda install absl or pip install absl,it all doesn’t work.
Hello , is there away to know the bounding box coordinates and crop it ?
Yes there is. You can edit the code in detect_video.py in order to perform this. look for the variable "boxes" as they are the bounding box coordinates. Then just opencv to crop!
The AI Guy it didn’t work with me, do you have the code for it ? Please
Amazing video mate .. Is it possible for you to do image segmentation as well ? Love your videos ...
For sure will try and do this. Thanks!
Amazing video. Thank you
Have you put Non Max Supression in your Github tutorial syntax?
Hello, I really like your in depth tutorials, they are very helpful.
I successfully followed your tutorial about training YOLOv3 custom model on Cloud, I downloaded the .weights file and convert it to .tf model following this tutorial but when I use it for on-video object detection I get this error:
raise ValueError("Shapes %s and %s are incompatible" % (self, other))
ValueError: Shapes (255,) and (42,) are incompatible
any ideas how to solve this issue ?
I think this has to do with the number of filters edited in the cfg file ( before training i changed the default value of filters (which is 255) to 42 as i am training the model on 9 classes of objects)
Same problem. Have you resolved it?
Hi saw your video and liked it a lot. I have a doubt:
How can I add the path of a video that I am getting in real time from ROS - its the real time camera feed of a drone in gazebo. How do I add its path?
great tutorial, just wondering, if i have to add multiple cameras inside the detection_video.py, what should i change in order for it to work? i know i can add cv2.Videocapture(0) and cv2.Videocapture(1). but im not too sure about everything else.
thanks for this very clear video :)
but i have 2 questions. 1.)how can we adjust position of output frame, half of it out of screen.
and 2.) i have this error : warning: Error opening file (/build/opencv/modules/videoio/src/cap_ffmpeg_impl.hpp:856) and also i can't write output video
please help me...
i am sorry i recognized that i don't have opencv-python. but after i did ' pip install', i have this error :
[ERROR:0] global C:\projects\opencv-python\opencv\modules\videoio\src\cap.cpp (392) cv::VideoWriter::open VIDEOIO(CV_IMAGES): raised OpenCV exception:
OpenCV(4.1.1) C:\projects\opencv-python\opencv\modules\videoio\src\cap_images.cpp:253: error: (-5:Bad argument) CAP_IMAGES: can't find starting number (in the name of file): 'C:/Users/User/Desktop/AIGUY_videos/Object_Detection_API_master/data/video/output.avi' in function 'cv::icvExtractPattern'
can you help me??
Can you create custom object detection using yolo V3 and open CV Step by Step
Yes I will try and make a video on this. I like it a lot.
I am also working on the same kind of the project I want to know how to create an alert wherever a certain object is detected?
Thanks a lot man. This was quite helpful.😁
thanks a lot buddy.. was very insightful.
Great video man!
Sir I have tried out a one line code similar to that could detect object on images ,on a video and saved it in similar way you did now but the problem was that I wasn't able to use that code for my custom yolo v3 detector, is there something I need to do?
I have made the custom yolo detector using your previous video❤️ in colab ,can you suggest any solution?
I show how to use custom weights file in this video, have you tried that way?
Hi, Thank you very much for an awesome video. Liked and Subscribed. There is a question I would like to ask. I saw in the label folder there is a coco text file with all the labels of the coco dataset. I try to reduce it to only certain items but it doesn't work. Could you suggest to me a way to do so?
Thanks for this video! Do you have any plans for object detection on a live rtsp stream from an IP cam?
Dan Danny i will definitely be looking to do a video on yolo for livestream video.
@@TheAIGuy Awesome! Thank you!
@@TheAIGuy when ?
How can I pause training, change the batch size, and resume training using the same loss curve?
how can we convert the detected object text into speech ?
Do let me know if you get how to convert into speech. Thanks
Can you share the paper/material on yolov3 tiny architecture? All I find is the standard yolov3 architecture only.
Hi when my trying to convert my custom weights(with classes 2) to tensorflow m getting this following error, can you please tell me the solution
File "C:\Users\hp\Desktop\Object-Detection-API-master\yolov3_tf2\utils.py", line 68, in load_darknet_weights
conv_shape).transpose([2, 3, 1, 0])
ValueError: cannot reshape array of size 75571 into shape (256,512,1,1)
I'm looking into this as someone else has just raised the same issue. Thank you and hopefully will be fixed soon!
@@TheAIGuy thank you as soon as you fix the issue or find any solution please post it
@@hrithikguntuka4300 Change the class in python script to your num of class
TAG, Another great video! I'm a big fan!
Questions on the '--video' webcam flag, if the '0' is replaced with and IP address ( to an IP camera) would the code still work? Thanks and looking forward to your next video.
Hey! I'm appreciate for your another great video. and Im having an error in the process of loading my custom weight file. Error message it "ValueError: cannot reshape array of size 73122 into shape (256,128,3,3) (yolov3-cpu)". I have set the command like
python load_weights.py --weights './weights/yolov3_custom_last.weights' --output './weights/yolov3-custom.tf' --num_classes 1. But getting this error. How should I fix this?
i'm facing the same issue..did you get a way to solve it ?
Try this command: python load_weights.py --weights ./weights/yolov3_custom_last.weights --output ./weights/yolov3-custom.tf --num_classes 1 --tiny; Also becase you did custom weight file, you need search num_classes in visual studio code and replace all num_classes = 1.
Can u please tell me how can I use a wireless camera to give an input.............please help
Can we add more custom objects detection to earlier yolov3 weights?
No unfortunately you need to retrain model with new classes from scratch. :(
@@TheAIGuy Ohh ok i thought so, thanks for the help.
When I give a video.mp4 (with audio) file for detection. The output file doesn't contain the audio..how to solve this issue..
Any possibility to train with more dataset on a custom trained model.
Yes you can train a custom yolov3 model and it will work just the same.
Can I implement it to mobile phone to detect pedestirans from phone's camera?
Is it possible to remove classes from the already pre-trained yolov3 model?? For example i want to only detect persons but not the other classes without retraining the model, is this possible?
It is not possible to remove classes from the train model, however you could add logic in the code to only show the detections for persons. This is the easiest way. You would just create an if statement and have it make sure the class is person before drawing detections.
tensorflow/core/platform/cpu_feature_guard.cc:142] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2
I'm getting this error
can i do that without Gpu ? i was train my model in cloud and i dont have GPU
Yes you can perform this video without GPU! Just run "pip install -r requirements.txt" instead of doing the conda commands.
Hi! How to run video from webcam on tiny-yolo? Couse when I run just yolo I get only 3-4FPS per second on GPU and that is not smooth.
Hello! Is there a way to insert the program into a raspberry Pi 3 Model B?
Brother thanks, how can I get the output (detected objects with time) in a table format. You are doing amazing.
how to set that it can only detect 1 object only like human or car only?
is it the same steps on a macbook too?
Why when I activate the CPU or GPU code
The program give me error : DLL load failed : the file cannot be accessed by the system
But when I do not activate any of those
The program run just fine with the GPU
????
What DLL is it missing?
@ابراهيم عبد الرزاق عبد الحسين hey, i'm also getting the same error. But can you please tell me how did you run it without activating cpu or gpu.
when i read image from webcam..it seems very slow..is there any way to solve this ???
Can you make a same tutorial using YOLO in Colab?
Can find yolov3 in the cloud videos already on my channel! But I will look at making Tensorflow YOLOv3 videos in Colab soon.
Hey, I wanted to deploy this model on a react website, like how can i do that? I was planning to deploy the website on AWS cloud.
could i run it in tensorflow-gpu==2.2.0 ?
Good lecture
hey .. great video .. is that possible to do this on google collab?
Hey
I am unable to find Custom_last_weights file
Can anyone help me in this especially @The AI Guy
License Plate recognition using YOLO ,please!
is cuda and cudnn free? even if you want to use that in a company?
Can it detect staircase?
No I don't think staircase is in the coco pretrained model unfortunately.
@@TheAIGuy Thanks for letting me know.
Hi The AI Guy, Can I feed in video from local serveraswell ?
What should I use for getting 30FPS? Should I use different model or Is there any other way to increase FPS in YOLO-V3?
If you use the yolov3-tiny model you can get over 30FPS quite easily. Try that and see if it meets your needs.
I want to say that waiting for your next video on Image Classification Android app using Tensorflow Lite. Please make the video soon. Thanks
does it work for YOloV4?
Hi, how do we get labels while we are detecting with camera, we are trying to build this app for blind people, it will detect objects and tell what it is so im trying to use your model and stuff now i need to get labels while real time detecting it. any help appriate it . respect
how can I get the model integrated with a Tello Drone?
Ill have to look into this for you!
Cool Stuff man !, I just want to ask why my video output saved from the webcam is so fast and too short, can you help me please?
Can i get workflow diagram for this project without api its urgent please help
can we implement this on windows 10 ?
will it work with CPU 2.2GHz and 8GB RAM this real-time webcam detection? I am planning on deploying ur code in a small shop. Also does it only work on ubuntu? can we run python executable from ubuntu in windows?
Is there any solution for gpu compute capability 3.0?
Eres grande, me ayudaste a salvar el semestre!
I have problem running load_weights.py with my custom weights. When I set the number classes which is 1 in my case I get the error: FATAL Flags parsing error: flag --num_classes='1': invalid literal for int() with base 10: "'1'". I tried to use another number I still get the error. Thank you in advance. Great videos!
can you reply with the exact command you are running! Once, I see it i'll be able to spot the issue.
off the top of my head though you probably set --num_classes '1', and for some reason its reading it as a string so try putting no quotes --num_classes 1
@@TheAIGuy You were right by not including the ' ' in the command also for the file path it worked. I'm using the Anaconda prompt not sure if that could be the reason. Thanks for the responds!
hi im using ubuntu ,when i run python load_weights.py im getting this error illegal instruction ( core dumped) any one can help me
why do i only have 2 fps with the cam ?
is it possible to run this on googles GPU?
First of all i wanna thank you so much for your great work and effort I really appreciate what you do❤️
I have a question, is this code work on google colab in the same way ?? And thank you❤️
when im trying to run this line "!./darknet detector train /content/darknet/data/obj.data /content/darknet/cfg/yolov3-custom2.cfg darknet53.conv.74 -dont_show" in colab it shows like this "/bin/bash: ./darknet: Is a directory". please help me
this is because you must be in the wrong directory. The command only works in the darknet directory so just run "%cd darknet" in the line above your training command and it should now work!
Thank u so much...it's working..! Keep going
How can I track people in real-time with web cam?
Where is Custom_last_weights filee..? My project is stuck now anyone please help me it's my college finall year project
Hi AI guy,
After converting yolov3_custom weights to tensorflow model, by running the below code -
python detect_video.py --weights ./weights/--yolov3-custom.tf-- -- video 0 -num_classes 2 --classes ./data/labels/obj.names
It shown me the below error at last:
raise ValueError("Shapes %s and %s are incompatible" % (self, other))
ValueError: Shapes (255,) and (21,) are incompatible
Please help.
Hello AI Guy,
Please assist me here
It is giving me an dll error plzz help
Anyone know how to fix this error: Shapes error.
File "C:\Users
ob26\anaconda3\envs\yolov3-gpu\lib\site-packages\tensorflow_core\python\framework\tensor_shape.py", line 1115, in assert_is_compatible_with
raise ValueError("Shapes %s and %s are incompatible" % (self, other))
ValueError: Shapes (255,) and (18,) are incompatible
how to add one more class on custom trained model?
Yeah I also want this
And I. Need to know how train the custom trained model with more dataset..
@@akhilgkrishnan4469 it is not possible to add just one class on top of previously trained model.
You need to retrain all classes again
Is possible to add more dataset to current trained model...
@@akhilgkrishnan4469 retrain your model again with new data ( old + new )
Hi, AI Guy kindly develop a fire detection system by using the yolov3 model with live stream
thanks Dude
Hello AI Guy. Great work helping everyone to learn AI. How can I connect with you for future, possible collaboration? Thanks, fanel
So I have successfully trained my own dataset using the training in the cloud video and have obtained my pretrained weights file. I ran the following command python load_weights.py --weights ./weights/yolov3_custom_last.weights --output ./weights/yolov3-custom.tf -num_classes 9 and recieved the folowing error: Traceback (most recent call last):
File "load_weights.py", line 34, in
app.run(main)
File "C:\Users\tronn\Anaconda3\envs\yolov3-gpu\lib\site-packages\absl\app.py", line 299, in run
_run_main(main, args)
File "C:\Users\tronn\Anaconda3\envs\yolov3-gpu\lib\site-packages\absl\app.py", line 250, in _run_main
sys.exit(main(argv))
File "load_weights.py", line 21, in main
load_darknet_weights(yolo, FLAGS.weights, FLAGS.tiny)
File "C:\Code\Object-Detection-API\yolov3_tf2\utils.py", line 68, in load_darknet_weights
conv_shape).transpose([2, 3, 1, 0])
ValueError: cannot reshape array of size 49334 into shape (256,128,3,3)
Looking into this for you.
The AI Guy thank you. I didn’t let the whole training Finnish in the cloud. Stopped it at about 2k something iterations, I did test it on the cloud model and it did work. Used those weights on the conda environment and got that error. Also- I only used 49 images as my original dataset amount. Another thing - could there be some sort of mismatch with the config file and the python script that creates the tensor?
The AI Guy and thanks man!
The AI Guy hey any chance you found what caused this error? I’m thinking of going the full Linux route to see if I can get this all working.
SO I gave in and went the Linux route. I was up until 3:00am in the morning but I have darknet and tensorflow all working perfectly. I am going to try to retrain my custom dataset in this environment now and will let you know how it went. Also, my problem might have been due toi an extra space in my classes.txt file. Going to make sure that is not the case.