Great work and explanation! I have a quick question: if one wants to change/update the object detection method, e.g.using Yolov5, can one simply declare the new one at the beginning and make sure to provide the right format in the main loop at: "boxes, scores, classes, nums = yolo.predict(img_in)"? I assume that, if these variable are correctly mapped, the rest of your tracking work should be directly portable to the new detection approach. Can you please let us know? Thanks again for the amazing work!
Thank you for the tutorial Please how do I display the number of count on YOLO v5? Thanks for the video I have a task which is pedestrian detection and congestion measurement I need to build a model that can detect pedestrian and also be able to say if the frame is congested For example let's say in a frame greater than 20 people means congested
The combination of Yolo and deep SORT make it super slow to run for it to be considered real time. I'd recommend centre point based tracking for higher fps
Which version of python do you use, pls ? I'm trying to lauch the process on windows but it doesn't work. I'm not able to download the old package version specified in your video.
Hi your video has given me new ideas I have 2 questions. First, can I also donate you a coffee via Paypal? Second, I may use parts of your code in my thesis? I am fascinated by your programming style and the efficiency of the code. Recommend your videos to all my friends :D
Hi Dennis, Please don’t worry about the Coffee, appreciate it. Please also free feel to use the code for your further work. Grateful that it’s useful. Thanks
hi awdhoot does given code is working? can you please help me in fixing this error? When i am running my code i got error generate_detection.py .. Traceback (most recent call last): File "object_tracker.py", line 31, in encoder = gdet.create_box_encoder(model_filename, batch_size=1) File "/content/drive/My Drive/Single-Multiple-Custom-Object-Detection-and-Tracking-master/tools/generate_detections.py", line 103, in create_box_encoder image_encoder = ImageEncoder(model_filename, input_name, output_name) File "/content/drive/My Drive/Single-Multiple-Custom-Object-Detection-and-Tracking-master/tools/generate_detections.py", line 84, in _init_ "net/%s:0" % input_name) File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/ops.py", line 3846, in get_tensor_by_name return self.as_graph_element(name, allow_tensor=True, allow_operation=False) File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/ops.py", line 3670, in as_graph_element return self._as_graph_element_locked(obj, allow_tensor, allow_operation) File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/ops.py", line 3712, in _as_graph_element_locked "graph." % (repr(name), repr(op_name))) KeyError: "The name 'net/images:0' refers to a Tensor which does not exist. The operation, 'net/images', does not exist in the graph."
encoder = gdet.create_box_encoder(model_filename, batch_size=1) this line is giving an error in goolge colab, The name 'net/images:0' refers to a Tensor which does not exist. The operation, 'net/images', does not exist in the graph Can you help me with this?
While running with yolov3-tiny getting this error how to resolve this please help File "C:\Users\windows\Documents\Single-Multiple-Custom-Object-Detection-and-Tracking\venv\lib\site-packages\tensorflow\python\framework\tensor_shape.py", line 1117, in assert_is_compatible_with raise ValueError("Shapes %s and %s are incompatible" % (self, other)) ValueError: Shapes (3, 3, 3, 32) and (3, 3, 3, 16) are incompatible
Thanks for video, i have a question about non_max_suppression. In YoloV3 there is already yolo_nms function. Why non_max_suppression is used in deep sort side again?
Thanks, it's a great video. Can i ask you, how can i get the frame when an object appear and disappear ? I want to track objects and save theirs history for my projects. Can you help me ?
no bounding box or detection after running the code, only video being played in the output when running on gpu but when runs on CPU detection is there in the output video. please help me out here
Hi sal ned, I am having the same issue here. Have you solved this issue? If no, can you please share with me how did you manage to run it with CPU? Will be really appreciated for the help.
This tutorial is awesome! I want to ask a question, there are a lot of warning error messages when i'm running convert.py. e.g. "WARNING:tensorflow:Entity could not be transformed and will be executed as-is." Is there any methods to solve this problem? Thank you so much!
Hi Meriem, In this video, it's with GPU because tensorflow has GPU implementation. However, opencv has non-GPU implementation. Please feel free to check this out - ua-cam.com/video/1LCb1PVqzeY/v-deo.html Hope it helps.
Thanks to you, it was very helpful to proceed with the project. Thank you But I have a question. I want to draw several lines and caculate the number of objects between the lines and the lines, but I am asking this question because it is not working well Here's the code I wrote height, width, _ = img.shape cv2.line(img, (0, int(3*height/6)), (width, int(3*height/6)), (0, 255, 0), thickness=2) cv2.line(img, (0, 0), (0, int(3*height/6)), (0, 255, 0), thickness=2) #1 sit left line cv2.line(img, (int(3*width/12), 0), (int(3*width/12), int(3*height/6)), (0, 255, 0), thickness=2)#1 sit right line or 2 sit left line cv2.line(img, (int(6*width/12), 0), (int(6*width/12), int(3*height/6)), (0, 255, 0), thickness=2)#2 sit right line or 3 sit left line cv2.line(img, (int(9*width/12), 0), (int(9*width/12), int(3*height/6)), (0, 255, 0), thickness=2)#3 sit right line or 4 sit left line cv2.line(img, (int(12*width/12), 0), (int(12*width/12), int(3*height/6)), (0, 255, 0), thickness=2)#4 sit right line what code should i add to operate normally
encoder = gdet.create_box_encoder(model_filename, batch_size=1) this line is giving an error in goolge colab, The name 'net/images:0' refers to a Tensor which does not exist. The operation, 'net/images', does not exist in the graph Can you help me with this?
To detect the direction, you can using coordinate of an object x, y in 1) current frame and 2) previous frame and then you can determine the direction For heat map, there are many ways to do that. I would suggest firstly you divide the frame into a grid (many tiny zone), then by counting how many objects inside each of the tiny zone.
is anyone getting this error? I get this with the encoder, by 32:00 ``` KeyError: "The name 'net/img:0' refers to a Tensor which does not exist. The operation, 'net/img', does not exist in the graph." ```
Hi eMaster, I have ran into a problem where there is no detection or bounding box at my output video. I have manually printed out the boxs, scores and classes, but these parameters only return values on the first frame while other remaining frames just remains an empty list. Does anyone here encounter such issue? Or did anyone manage to solve it? I am aware that I only generated the 3 files when converting yolo to tensorflow: checkpoint, yolov3.tf.data-00000-of-00001, yolov3.tf.index. Could this be the problem? If yes, does anyone has experience with this?
The combination of Deep sort and Yolo have this issue. I've tried several implementation of this with the same low fps. Id recommend other tracking algorithms instead of deep sort
However, I can see that several person's track_id are switched when they cross each other. 1:15:33 (top left ID 7 and ID 60) 1:15:44 (top right ID 5 and ID 2) Why do you think that this problem happens? Can we save more tracking path to recognize whether the two persons are not switched?
Can onyones tell me how can I change the code from the cocodatabase if I only need few of the classnames and not the all 80? I didn't found the result in Web. Plz help :D
You can simply set it with "if statement" please find some similar ideas here - ua-cam.com/video/zi-62z-3c4U/v-deo.html You can change the class_name over here to the types that you want. --> if class_name == '': Hope it helps.
I have trained yolov3 using custom dataset. BUt tracking gives the error: raise ValueError("Shapes %s and %s are incompatible" % (self, other)) ValueError: Shapes (3, 3, 3, 32) and (3, 3, 3, 16) are incompatible Can you please look tell how to solve it
Suppose i want to do number plate extraction .. For every frame, it will give number details, i don't want so many times.. I want 1 time extraction for 1 car.. How to do it sir.. Please upload your car video in your repo
One more question , if i just want to detect the motorbike, and get the number of motorbike through the line. I think i dont need to show the frame of video, because it can reduce the work of gpu, and i will have a result of a video quickly. That is my thinking. Can you check that? If it is right, so how can i change the code? Thank you so much.
I was thinking it would be two steps. Firstly, you need to detect the motorbike's nameplates and capture the pictures. And then you can use text recognition, like pytesseract, to recognize the text from the saved pictures.
@@thongphamminh3572 hello, i am also trying to do a similar project, i just wanted to ask whether can you help me by sharing your code for reference, thank you
Hi friend, Can you help me with the step 3: Execute convert.py in a terminal? I store the files in the motorbike detection (C:\Users\Administrator\PycharmProjects\motorbike detection\motorbike detection).but I cannot find the yolov3_deepsort(i have download your files in github). And one more problem is that when i command python convert.py in terminal, this is what i receive (Python was not found; run without arguments to install from the Microsoft Store, or disable this shortcut from Settings > Manage App Execution Aliases.) Can you also help me this problem?
And How can i increase the fps in the detection video? I think i can use gtx 1080 ti to achieve, but i am not sure about that. Thank you for your help and video.
Yolov3 and Yolov5 are able to do that. If an object is human, opencv or mediapipe can also serve the purposes, like bodypose, face detection, etc. Please feel free to check out my yolov5 tutorial. It allows you to detect object via webcams, videos or even real-time streaming UA-cam.
I am facing an error at the following lines: encoder = gdet.create_box_encoder(model_filename, batch_size=1) the error is : NotFoundError: model_data/mars-small128.pb; No such file or directory I tried to change the directory(cd) to 'deep_sort' as well, but still got the same error. Can any one please help me?
@@teezhenquan5855 thank u. I figured it out where the problem was. Actually there is a space after 128 in the file name mars-small128.pb. I was missing this which caused the error
hello, please i need to implement this with yolov4 and yolov4 tiny, please i need help how can i convert the darknet model to tf2 format and how am I going to load the model, thank you.
So I'm having some trouble converting a custom darknet weights file to a .tf format. I'm getting a value error: cannot reshape array of size 269633 into shape (256,128,3,3). I'm not sure how to fix this.
@@rabinsubedi7669 sorry dude, I wasn't able to fix that. Though I think the error occurs when you change the number of classes your custom dataset would need to sort through. Not sure what the hell is causing it.
@@rabinsubedi7669 so I retraced my steps. Turns out it has something to do with darknet configuration. I had to change the number of classes before extracting weights from a custom training set, then make sure I have the classes correctly indexed by whatever annotation tool I'm using. For some reason, when I did it like that - it worked. I also moved everything to cloud instead of using my local machine - so maybe that has something to do with it?
I cannot detect objects far from the camera even if they are clear (I detect when I bring the object closer), is there an optimal distance or config to adapt?Thanks
@@bellagofarsah8735 CUDA is exclusive built for the NVIDIA architecture, in this moment there is no possibility to use CUDA having a AMD gpu, an alternative would be to use Google Collab and set the GPU environment in order to take advantege of a GPU in the detection task.
Can we use a custom model yolov3, i want to track object, but with my custom dataset. Can you show me how? And thanks a lot. Your video is very helpful.
Hi. I would like to get in touch with for a future project on our company. Can you please send me an email with your contact, so we can schedule a call ? Thanks
It's really a great tutorial video. Thank you so much!
Thanks for watching and your kind support.
Thanks, another great implementation course!
Awesome!! Thanks man
Thanks for making detailed video on this topic ... Thank you so much
Wonderful content and great explanation. Huge respect
Thanks it is very helpful , can you please upload some videos on detectron2
Great work and explanation!
I have a quick question: if one wants to change/update the object detection method, e.g.using Yolov5, can one simply declare the new one at the beginning and make sure to provide the right format in the main loop at: "boxes, scores, classes, nums = yolo.predict(img_in)"?
I assume that, if these variable are correctly mapped, the rest of your tracking work should be directly portable to the new detection approach.
Can you please let us know? Thanks again for the amazing work!
Just wonderful
Thank you for the tutorial
Please how do I display the number of count on YOLO v5?
Thanks for the video
I have a task which is pedestrian detection and congestion measurement
I need to build a model that can detect pedestrian and also be able to say if the frame is congested
For example let's say in a frame greater than 20 people means congested
thanks for this wonderful course ...... is this applicable when using yolov7 or there will be some differences?
The combination of Yolo and deep SORT make it super slow to run for it to be considered real time. I'd recommend centre point based tracking for higher fps
getting following error
cv2.resizeWindow('output', 1024, 1024)
cv2.error: OpenCV(4.1.2) /io/opencv/modules/highgui/src/window_QT.cpp:592: error: (-27:Null pointer) NULL guiReceiver (please create a window) in function 'cvResizeWindow'
Which version of python do you use, pls ? I'm trying to lauch the process on windows but it doesn't work. I'm not able to download the old package version specified in your video.
Thank you! It's so interesting.
Hi your video has given me new ideas I have 2 questions.
First, can I also donate you a coffee via Paypal?
Second, I may use parts of your code in my thesis?
I am fascinated by your programming style and the efficiency of the code.
Recommend your videos to all my friends :D
Hi Dennis,
Please don’t worry about the Coffee, appreciate it.
Please also free feel to use the code for your further work.
Grateful that it’s useful.
Thanks
you did well, appreciate it!
Can I count peoples in first video, tracking and detection is good, but can I also count they ...
Great Work.....
One Query.... How to integrate Faster RCNN object detection with deep SORT
Even with tiny-yolo the average fps is 14. I recommend not using FRCNN for tracking as speed is very low, around 2-3 fps
Thank you for great Tutorial,
I Want to Ask if This Tutorial can work similarly on Nvdia jetson ?
I got all the packages installed, it takes in the test video but doesnt create any boxes, Im confused pls help!
is your weights folder contain 00-01 and 01-02 data?
Really great video just a small query
Can we increase frame rate ??
If yes then how ??
hi awdhoot does given code is working?
can you please help me in fixing this error?
When i am running my code i got error generate_detection.py
..
Traceback (most recent call last):
File "object_tracker.py", line 31, in
encoder = gdet.create_box_encoder(model_filename, batch_size=1)
File "/content/drive/My Drive/Single-Multiple-Custom-Object-Detection-and-Tracking-master/tools/generate_detections.py", line 103, in create_box_encoder
image_encoder = ImageEncoder(model_filename, input_name, output_name)
File "/content/drive/My Drive/Single-Multiple-Custom-Object-Detection-and-Tracking-master/tools/generate_detections.py", line 84, in _init_
"net/%s:0" % input_name)
File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/ops.py", line 3846, in get_tensor_by_name
return self.as_graph_element(name, allow_tensor=True, allow_operation=False)
File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/ops.py", line 3670, in as_graph_element
return self._as_graph_element_locked(obj, allow_tensor, allow_operation)
File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/ops.py", line 3712, in _as_graph_element_locked
"graph." % (repr(name), repr(op_name)))
KeyError: "The name 'net/images:0' refers to a Tensor which does not exist. The operation, 'net/images', does not exist in the graph."
@@robo2.069 haven't tested this code but you can get same code from aiguyscode github its working fine
@@awdhootkanawade okay tanku but this code also works but with some changes in code.
encoder = gdet.create_box_encoder(model_filename, batch_size=1) this line is giving an error in goolge colab,
The name 'net/images:0' refers to a Tensor which does not exist. The operation, 'net/images', does not exist in the graph
Can you help me with this?
While running with yolov3-tiny getting this error how to resolve this please help
File "C:\Users\windows\Documents\Single-Multiple-Custom-Object-Detection-and-Tracking\venv\lib\site-packages\tensorflow\python\framework\tensor_shape.py", line 1117, in assert_is_compatible_with
raise ValueError("Shapes %s and %s are incompatible" % (self, other))
ValueError: Shapes (3, 3, 3, 32) and (3, 3, 3, 16) are incompatible
did u solve it?
Thank you so much. Can you make a video for YOLO V4 too ?
Thanks for video, i have a question about non_max_suppression. In YoloV3 there is already yolo_nms function. Why non_max_suppression is used in deep sort side again?
thank you
Thank you for an awesome video.
Is there a way to assign to assign the vehicle number plate as the object id for tracking.
Thanks, it's a great video. Can i ask you, how can i get the frame when an object appear and disappear ? I want to track objects and save theirs history for my projects. Can you help me ?
I would like to transfer this information to a 2D map. Is this possible?
no bounding box or detection after running the code, only video being played in the output when running on gpu but when runs on CPU detection is there in the output video. please help me out here
Hi sal ned, I am having the same issue here. Have you solved this issue? If no, can you please share with me how did you manage to run it with CPU? Will be really appreciated for the help.
This tutorial is awesome! I want to ask a question, there are a lot of warning error messages when i'm running convert.py. e.g. "WARNING:tensorflow:Entity could not be transformed and will be executed as-is." Is there any methods to solve this problem? Thank you so much!
thanks a lot for the course, but i have a basic question, if my computer is cpu, can i run this code ? thank you
Hello emaster, I was thinking how to apply it for social distancing, any idea?
Thanks a lot. It was a great effort. How do I save the tracked object ID and corresponding trajectory position in a csv file?
how do i get the data for walking person detection thank you
KeyError: "The name 'net/images:0' refers to a Tensor which does not exist. The operation, 'net/images', does not exist in the graph." any help !!!
Dowgrading the tensorflow solved the issue: uninstall tf and install tensorflow 2.1.0
pip install tensorflow==2.1.0
Hey, great video! How can I cout vehicles when there isnt only up and down movement?
Thank you!
hi ,Thanks for your course. Can you please upload the example video ./data/video/testvideo.mp4 . The good test mp4 file is not easy to have, thanks
Hello ,is it necessary to have GPU to implement this course or not ?
Hi Meriem,
In this video, it's with GPU because tensorflow has GPU implementation.
However, opencv has non-GPU implementation. Please feel free to check this out - ua-cam.com/video/1LCb1PVqzeY/v-deo.html
Hope it helps.
@@eMasterClassAcademy yes i already try this video and it was a great help thank you very much.
How can i speed the procces? I use 950m as GPU and it uses only %5 of it. fps rate is about the 4 . What should i have done for real time detection ?
Thanks to you, it was very helpful to proceed with the project. Thank you
But I have a question. I want to draw several lines and caculate the number of objects between the lines and the lines, but I am asking this question because it is not working well
Here's the code I wrote
height, width, _ = img.shape
cv2.line(img, (0, int(3*height/6)), (width, int(3*height/6)), (0, 255, 0), thickness=2)
cv2.line(img, (0, 0), (0, int(3*height/6)), (0, 255, 0), thickness=2) #1 sit left line
cv2.line(img, (int(3*width/12), 0), (int(3*width/12), int(3*height/6)), (0, 255, 0), thickness=2)#1 sit right line or 2 sit left line
cv2.line(img, (int(6*width/12), 0), (int(6*width/12), int(3*height/6)), (0, 255, 0), thickness=2)#2 sit right line or 3 sit left line
cv2.line(img, (int(9*width/12), 0), (int(9*width/12), int(3*height/6)), (0, 255, 0), thickness=2)#3 sit right line or 4 sit left line
cv2.line(img, (int(12*width/12), 0), (int(12*width/12), int(3*height/6)), (0, 255, 0), thickness=2)#4 sit right line
what code should i add to operate normally
ex) if class_name == 'car':
current_count += 1
if center_y
encoder = gdet.create_box_encoder(model_filename, batch_size=1) this line is giving an error in goolge colab,
The name 'net/images:0' refers to a Tensor which does not exist. The operation, 'net/images', does not exist in the graph
Can you help me with this?
How can we evaluate this model?
what are the software and hardware requirements
Can you guide me in how to detect an object using kinect xbox 360 camera
hi. im having issue where no bounding box/detection occurs for video. using the yolov3. is it because when i run convert.py , it only produce 3 files?
How can we detect the direction on moving object?
How to create heat map?
Thankyou so much for this video!!!!!!!!
To detect the direction, you can using coordinate of an object x, y in 1) current frame and 2) previous frame and then you can determine the direction
For heat map, there are many ways to do that. I would suggest firstly you divide the frame into a grid (many tiny zone), then by counting how many objects inside each of the tiny zone.
Can you send me this code colab file please
@@sanabatool5573 Hey sana can you get your answer? I want to Colab notebook code
@@karthickraja1026 no. Now i am using YOLOv5.
@@sanabatool5573 Hey sana can you give me some advice ? I want to use yolov5 too , please!
could you please explain (1,100) in the return value of yolo
Could you please tell what are the evaluation metrics for deep sort? How can I measure the performance of this model?
good question, Also I am wondering performance measurement of my models. Do you have any ideas/suggestions?
is anyone getting this error?
I get this with the encoder, by 32:00
```
KeyError: "The name 'net/img:0' refers to a Tensor which does not exist. The operation, 'net/img', does not exist in the graph."
```
Hi eMaster, I have ran into a problem where there is no detection or bounding box at my output video. I have manually printed out the boxs, scores and classes, but these parameters only return values on the first frame while other remaining frames just remains an empty list. Does anyone here encounter such issue? Or did anyone manage to solve it? I am aware that I only generated the 3 files when converting yolo to tensorflow: checkpoint, yolov3.tf.data-00000-of-00001, yolov3.tf.index. Could this be the problem? If yes, does anyone has experience with this?
did u solve this issue?
tensorflow/compiler/jit/xla_gpu_device.cc:99] Not creating XLA devices, tf_xla_enable_xla_devices not set
help me plz
Hi, during the detection for person, can we detect when someone fall?
Great video, we appreciate it. I ran into one problem which was very low FPS of 2 of the output data. Is there a way to increase the FPS to 6 or 7?
could you share more details about your system on which you implemented.
Are you worked in cpu?
The combination of Deep sort and Yolo have this issue. I've tried several implementation of this with the same low fps. Id recommend other tracking algorithms instead of deep sort
ImportError: cannot import name 'convert_boxes' from 'yolov3_tf2.utils' . Please Help!!!
hi sir multiple object dtection and tracking using lstm and yolo we need so pls help
Hello. I notice that there is a pre train file .pb
Is there a way that I can custom train this file ? Please give me some advice.
Best regards.
did you manage?
how can i calculate the tracking accuracy and precision
However, I can see that several person's track_id are switched when they cross each other.
1:15:33 (top left ID 7 and ID 60)
1:15:44 (top right ID 5 and ID 2)
Why do you think that this problem happens?
Can we save more tracking path to recognize whether the two persons are not switched?
Same here. Did you solve the problem?
@starrysoon848 any progress?
1:07:00 tracking
Hi there, how can I test the accuracy of this system using a dataset with annotations?
Can onyones tell me how can I change the code from the cocodatabase if I only need few of the classnames and not the all 80? I didn't found the result in Web. Plz help :D
You can simply set it with "if statement"
please find some similar ideas here - ua-cam.com/video/zi-62z-3c4U/v-deo.html
You can change the class_name over here to the types that you want.
--> if class_name == '':
Hope it helps.
@@eMasterClassAcademy Thank you so much for the reply I am so grateful. It also worked immediately.
I don’t have gpu , can I implement this on cpu ?
I have trained yolov3 using custom dataset. BUt tracking gives the error: raise ValueError("Shapes %s and %s are incompatible" % (self, other))
ValueError: Shapes (3, 3, 3, 32) and (3, 3, 3, 16) are incompatible
Can you please look tell how to solve it
did u fix it?
where u get deepsort repo end me link
this can using google colab?
Thanks for watching.
Yes it works with google colab.
Please feel free to check out my Yolov5 for more latest version.
Enjoy~
Suppose i want to do number plate extraction ..
For every frame, it will give number details, i don't want so many times..
I want 1 time extraction for 1 car..
How to do it sir..
Please upload your car video in your repo
Well its simple, just take the object id and make sure it isn't there in your license plates list.
hello, i just wanted to say i am also doing the similar project and ask you whether you can share your code for reference. thank you
Do the program detect motorbike? If yes, can you show me how to change the code? Thank you.
One more question , if i just want to detect the motorbike, and get the number of motorbike through the line. I think i dont need to show the frame of video, because it can reduce the work of gpu, and i will have a result of a video quickly. That is my thinking. Can you check that? If it is right, so how can i change the code? Thank you so much.
I was thinking it would be two steps. Firstly, you need to detect the motorbike's nameplates and capture the pictures. And then you can use text recognition, like pytesseract, to recognize the text from the saved pictures.
@@eMasterClassAcademy Thank you
@@thongphamminh3572 hello, i am also trying to do a similar project, i just wanted to ask whether can you help me by sharing your code for reference, thank you
Can we implement it into a grid for heat map? Thanks
Kindly do it with Yolov5 as well
How can I execute everything in Google Colab because the resources I have right now won't be sufficient for the project.
Check out THE AI GUY's yolov4-deepsort repo
Hi friend, Can you help me with the step 3: Execute convert.py in a terminal?
I store the files in the motorbike detection (C:\Users\Administrator\PycharmProjects\motorbike detection\motorbike detection).but I cannot find the yolov3_deepsort(i have download your files in github).
And one more problem is that when i command python convert.py in terminal, this is what i receive (Python was not found; run without arguments to install from the Microsoft Store, or disable this shortcut from Settings > Manage App Execution Aliases.) Can you also help me this problem?
And How can i increase the fps in the detection video? I think i can use gtx 1080 ti to achieve, but i am not sure about that.
Thank you for your help and video.
Sir, i want to make object detection in real time live streaming with camera
Can u suggest any study purpose video to me
Yolov3 and Yolov5 are able to do that. If an object is human, opencv or mediapipe can also serve the purposes, like bodypose, face detection, etc.
Please feel free to check out my yolov5 tutorial. It allows you to detect object via webcams, videos or even real-time streaming UA-cam.
Can we runit in VS Code
How do I get the videos you have used?
I want them for my set of codes too, Thanks a bunch!
01:02:43
Oh and what python version are you using? I tried but I could not get a output
For suppose, if we shift to a different camera, can we track the same object with the same ID?
have you done this part?
I am facing an error at the following lines:
encoder = gdet.create_box_encoder(model_filename, batch_size=1)
the error is :
NotFoundError: model_data/mars-small128.pb; No such file or directory
I tried to change the directory(cd) to 'deep_sort' as well, but still got the same error.
Can any one please help me?
For the first issue, You can try this: github.com/theAIGuysCode/yolov3_deepsort/issues/22
@@teezhenquan5855 thank u. I figured it out where the problem was. Actually there is a space after 128 in the file name mars-small128.pb. I was missing this which caused the error
hello, please i need to implement this with yolov4 and yolov4 tiny, please i need help how can i convert the darknet model to tf2 format and how am I going to load the model, thank you.
Would you consider using yolov5? It can directly be implemented with PyTorch, and you can build your functions on top of the detect.py easily.
My Laptop 8GB RAM Utilization goes to 98%...i am using webcam. it seems that this solution only fits to GPU, right?
Well, It should be fine
ModuleNotFoundError : No module named 'yolov3_tf2'
This error shows. How can I solve this issue ?
pip install yolov3_tf2
how do i make it run on gpu (cuda)
So I'm having some trouble converting a custom darknet weights file to a .tf format. I'm getting a value error: cannot reshape array of size 269633 into shape (256,128,3,3).
I'm not sure how to fix this.
hello @twinblade02 I got the same error. Were you able to fix this error? If yes, I would be grateful if you could let me know how its done. Thanks!
@@rabinsubedi7669 sorry dude, I wasn't able to fix that. Though I think the error occurs when you change the number of classes your custom dataset would need to sort through. Not sure what the hell is causing it.
@@twinblade02 Thanks for the reply man. I actually used the weight file from custom training in another repo (for eg: AI Guy repo) and it worked.
@@rabinsubedi7669 so I retraced my steps. Turns out it has something to do with darknet configuration. I had to change the number of classes before extracting weights from a custom training set, then make sure I have the classes correctly indexed by whatever annotation tool I'm using. For some reason, when I did it like that - it worked. I also moved everything to cloud instead of using my local machine - so maybe that has something to do with it?
@@twinblade02 maybe; I am still working on local machine and it is not working.
I cannot detect objects far from the camera even if they are clear (I detect when I bring the object closer), is there an optimal distance or config to adapt?Thanks
Have you done in CPU or GPU?
@@karthickraja1026 CPU
@@elmehdiaa2882 What is a tensorflow version used?
@@karthickraja1026 PyTorch YoloV5
@@elmehdiaa2882 Are you working on total count objects?
how can we do it on ssd mobilenet model?
there's no data ?
I would like to do all the work with jupyter-notebook using your codes but it doesn't work.
Can you help my please?
pip install p2j for any script
p2j script.py in ( in cmd)
Thanks! I will try it inshaa Allah.
1:12:40 Closer thicker and farther thinner **
Please upload its colab file
just load all files in drive and mount drive and change the path to the object detection file path and write in cell--!python convert.py and so on.
@@tamzidalam8766 I got the same error, do you have answers?
@@tamzidalam8766 it will work if you change the code in generations_detections.py as "%s:0" in both input and output var under ImageEncoder class.
bro i dont have nvidia gpu i am having amd so what can i do to setup local gpu environment
me too bro, im using radeon graphic card, can someone tell us how to setup the first step of the video? thanks in advance
@@bellagofarsah8735 CUDA is exclusive built for the NVIDIA architecture, in this moment there is no possibility to use CUDA having a AMD gpu, an alternative would be to use Google Collab and set the GPU environment in order to take advantege of a GPU in the detection task.
Hello, great tutorial but I have got " unable to resolve "yolov3_tf2.models" intellisense may be missing" error. Any idea to solve this issue.
Convert.py is working fine. But yolov3.tf is available. How to get that
Same problem here
Can we use a custom model yolov3, i want to track object, but with my custom dataset. Can you show me how? And thanks a lot. Your video is very helpful.
@Mohamed KHENAFIF hi, not yet.
@@ucvuongminh2478 hello, i just wanted to ask did you complete with your custom data and real time yolo detection? thank you
better to use docker
3 commercials in 10 mins? wtf
한글버전 주세요
그냥 pycharm으로 할수 있어요!ㅎㅎㅎ
@@dkibe6006 혹시 파이팜으로 안켜져서 그런데 따로 미리 설정해야할게있나용??
Do you have an email I have a project for you please
Hi. I would like to get in touch with for a future project on our company. Can you please send me an email with your contact, so we can schedule a call ?
Thanks
Please feel free to drop an email to emasterclassenquiry@gmail.com
Happy to connect.
Thanks for another great course!