Yes i do want to know thank u sir

How can I use the GPU instead of cpu for running this

Hello Mr. Rob , I have watched many videos of yours just to try to make this work, I already installed linux ubuntu, Anaconda, cloned yolov5 and installed the requirements successfully, then when coming to the command : lc -tr I dont get the same result as yours, I only get this result of three lines: Mono license compiler copyright (c) 2009 by remobjects software No target/complist passed Have been working for 10 hours and I am still stuck in the first 3 minutes, can you please help me

same, have you found the solution? @@Hashemaverick

I remember asking chatgpt and he solved it easily @@VuAnh-ey3bi

Thanks for the sub! Spread the word :D

I love feedback like this Chris- it motivates me! So thanks for sharing it. Congrats on starting learning python, you’re going to love it.

Glad it helped!

Thanks for watching! Tell your friends 😊

WARNING! ELON MUSK SHOULD NOT USE THIS MODEL TO MAKE SELF DRIVING CARS!!! :D Just kidding (but also not kidding). Remember the video at that timestamp what when we made the detection confidence threshold very low and the IoU threshold very high - so it will over predict a lot of false positives. Thanks for watching!

Great question. Honestly I’m not the best person to answer this question. Like most things you can only optimize for speed up to the limitations of the hardware. Usually if I’m running out of gpu ram then batch size needs to be lowered. The issue with VRAM might be unavailable because the images need to be loaded into memory before predicting. What type of images are you running on? I’ve run on 720p with fairly large batch sizes and no issues.

Great point! At the time I released this yolov5 was the most commonly used. Even though it’s named yolov7 the open source version is not made by the same people who created v5. I want to eventually make a video showing the v7 version.

Thank you Rob@@robmulla for letting me know this. I look forward to learning the differences in working with v7 and whether it can deliver significantly better results. This assumes we can expect v7 to receive updates and maintenance. Otherwise using v5 may be the best option if it's the main version and better supported and much more commonly used. I think v7 should be renamed to YOLO-derivative-v7.

That's a good question. You could certainly modify the code to aggregate the labels, or use the output of the txt file to post process. I'm not sure of the best way to add metadata to video though.

Just a mere mortal here! Hope you find the video helpful.

You’re welcome 😊

@@robmulla A suggestion: You are very good at teaching. But this tutorial is little too fast. I had no idea when the terminal shows ' git not recognized as internal or external command. Then after long time I installed the git software. If you have mentioned the git installation that would have helped us alot. And at some point the video is going very fast. So please keep it in mind for future videos. But still u are a teacher for me, thankyou so much for the video....❤️❤️❤️

@@dani9609 I really apprecaite the feedback! I didn't think about how that might be confusing but am happy you mentioned it. Maybe I should make a different video about git. I'll try to do better in the next one.

Thanks! The default model is trained on the COCO dataset with some common labels. We can train a model that predicts custom objects. I will make a video soon that shows how that can be done.

@@robmulla perfect!

Thanks for watching! I also have a video on custom object detection with yolov7! ukposts.info/have/v-deo/ioiTqImukamgpok.html

Thanks for the feedback. Yes it’s an option in the detect.py file to save the results to a text file. You just need to set the flag.

Yes, this is totally possible but would require working with the base code. If you look in the detect.py the detections are saved as "det" github.com/ultralytics/yolov5/blob/master/detect.py#L136 The code currently shows the box with detections but you could modify it to display just a box with text for the objects detected. Hope that helps.

@@robmulla Thanks Alot Rob, You're an amazing human. Its a request when ever you get enough time try to make a video on how to write something in live frame as I described you before because I've search alot this topic on internet but I didn't get any right answer or direction for this issue. I'm so thankful to you.

4:41 in the arguments, see --save-txt. Saves the results '.txt'

Thanks for the feedback. You might need to `pip install yaml` if the module is not found it usually needs to be pip installed. Hope that helps.

That's great that you've trained your own model. I plan to make a video about the training process in the future. The problem you talk about is pretty common. What people sometimes do is train a two-stage detector. The first detector gets an idea for the scale of the objects and then the second predicts on the rescaled version. Also, when training you can augment your labeled images to be varying sizes so yolo doesn't overfit to the large signs. Of course if the sign is extremely small the model will always have a difficult time detecting. Hope that helps.

I'm not sure about the audible part, but you could easily make a python program that checks a video stream to see if people are detected. Playing the sound would depend on your operating system. Check out the top response to this SO post: stackoverflow.com/questions/16573051/sound-alarm-when-code-finishes

You’re welcome!

I can used this code for text detection !!!!

Absolutely! People are doing it already. Although it would require learning a lot more than what I'm showing in this 10 minute video :D

You could probably do that. You might want to check out my other video on training a custom object detector using yolo. Good luck!

Thanks for the feedback Richard. What do you mean by running from a script exactly. The code that I ran directly from the yolov5 repo is essentially a script. I was thinking that depending on how popular this video is I could make follow up videos showing how to train yolov5 on a custom dataset and applying it so that the prediction boxes are stored.

@@robmulla When I said python script I meant implementing YOLO detection in custom script that did a triggered operation depending on the detection/labels. if label = 'car' save frame as image with label as filename, or trigger something else to occur like turn on outside lights.

Thats a great idea. I'll have to think about how that would be implemented but is definately doable.

Great question. I didn’t go into detail about them but implementations of them can be found on GitHub or in the yolo source code.

I don't believe yolo can change the size of the video- but it can be done with cv2 in python. Look up "opencv python resize video and modify speed" speed is determined by FPS. Hope that helps.

Great questions @Vislone you are on the right track. 1. The base model was trained on the default COCO labels. Just google "COCO label list". You can however, train a model on anything if you have enough labeled images. I plan to still make a video about that. 2. Yes, the bounding boxes can be writted to a file OR you can go into the actualy detect.py and see how the code processes and store them in a different way. Good luck!

@@robmulla Thank you Sir :)

Thanks for watching. Great question. To make the model better at predicting specific objects it’s best to train or “fine tune” the model on a dataset of additional labeled images. I plan to make a video about this process. In your example you would need to train on a giraffe or squirrel specific dataset.

there are other libraries that specialize in text recognition such as tesseract, but I am sure you could use yolo as well.

Wow, that sounds like a really cool project. Definately outside my knowledge with regards to object detection. Maybe look into something like this? www.analyticsvidhya.com/blog/2021/12/step-by-step-guide-to-build-image-caption-generator-using-deep-learning/

Thanks for watching! That's a very specific request :D I can't garuntee I will do that but have you watched my other video about custom training. You might find it helpful: ukposts.info/have/v-deo/ioiTqImukamgpok.html

Thank you for the reference…will look into it

替换数据集即可。

Welcome to the wonderful world of python! I would reccomend starting with installing anaconda first from www.anaconda.com/ After that is installed, if you are running on windows, you can load the terminal by searching for the "anaconda prompt". If you are running a mac you can search for "terminal". Hope that helps!

I m using windows and using python 3.10 is it ok can plse do from this

I'm glad you are excited. I am using a linux machine and running everything in the command line. It should be similar with windows but I'm not as familiar with how to do it. Try using powershell and installing anaconda first. Good luck!

If i used this in cmd power shell will i get the output too??

Edit: i have used but the lc -tr to read the files is not working

@@user-uz6bf7op4f I think so

when running detect.py you can add the parameter "--save-txt" which will save the output into a text file. The first column in that file will be the class labels associated with the COCO dataset: tech.amikelive.com/node-718/what-object-categories-labels-are-in-coco-dataset/

@@robmulla no I mean that I am using the object detection model on my laptop jupyter notebook(offline) and I want to access the class name so that I can further send the data to other areas For example if the object detected is Car then "Car" class name would be stored into a variable,(let's assume we have to store class name in variable x ) so what will be the code for that And Sir thanks for the reply

Thanks! I think?

Thanks!

Probably not great for text but I have a video for that! Check it out here: ukposts.info/have/v-deo/p6migpqamYGZxqs.html

Hmm. You might need to train a custom detector. But something like that can be hard because it involves more that just detection and the model needs to know the surrounding contex in the image.

@@robmulla Wow!!!!! I never thought that you would reply back. Thnx a lot for your helpful suggestion.

Glad you liked the video. It may be slow if you don’t have a GPU. You can try running it in a Kaggle notebook - it might be faster.

Not sure what you mean. You would need the camera to be a source on your PC

Mate terminal in ubuntu

Hey. Thanks for watching the video. I know someone else mentioned the same issue. I see there is an active discussion about it on the yolo github page, you might want tor read that here: github.com/ultralytics/yolov5/issues/9844

@@robmulla THANKS A LOT!!!!!!. I got it fixed. Once again thanks for helping me out.

Absolutely! You just need a training dataset.

Interesting idea. I’ll have to think about it.

Sorry to hear that. Maybe paste the error message here? Are you using anaconda? Another way to test it out without having to do the setup would be to use something like google colab or kaggle notebooks - that will have everything preinstalled (except yolo). Those won't work with a local webcam though. Usually when I get errors I copy and paste them into google and 99% of the time someone has posted about the same problem on stack overflow.

@@robmulla okay, switching over to google colab. i don't need to use it on a webcam right now anyway. so all i do is clone the repo in the beginning, and then do i try to install the requirements.txt file? cuz it's not working in colab.

I don’t think that YOLO would be the best thing to use for this unless you are looking for specific text like a STOP sign. You might want to look into OCR techniques.

This is mate terminal in ubuntu. I have a whole video on my setup you should checkout. ukposts.info/have/v-deo/jJSTma-ki5mKunk.html

Thanks! Are you running on a system that doesn’t have a monitor directly connected like a remote server? You might need to disable the image output and instead store the results to a file using the appropriate flags. Hope that helps.

@@robmulla it's all a laptop that i am using. The objects are being detected ,it's just that the video isn't playing.Anyways thanks for helping me out.🙂

Great question. I actually have a video that discusses this in detail. Check it out here: ukposts.info/have/v-deo/ioiTqImukamgpok.html

@@robmulla this is perfect, i really needed this. Thanks a lot.

It must not be installed correctly or you are trying to run from a different environment, because otherwise it should be found when importing. Try running your code in a kaggle notebook to double check maybe?

I had this error, and I run it with python 3.10 instead of 3.11, you should try this

@@theophanedebellabre2676 ok thanks I would for sure

When running detect.py use the --save-txt flag to save the results in a text file.

Hey Alamin. I don't use pycharm but the IDE you use shouldn't impact your ability to run this code. Good luck!

Thanks for the feedback. I’ve never seen that one before. Could it be an issue with the file name? Try renaming it to something simple. Good luck

detect: weights=yolov5s.pt, source=0, data=data/coco128.yaml, imgsz=[640, 640], conf_thres=0.25, iou_thres=0.45, max_det=1000, device=, view_img=False, save_txt=False, save_conf=False, save_crop=False, nosave=False, classes=None, agnostic_nms=False, augment=False, visualize=False, update=False, project=runs/detect, name=exp, exist_ok=False, line_thickness=3, hide_labels=False, hide_conf=False, half=False, dnn=False, vid_stride=1 YOLOv5 🚀 v6.2-228-g6ae3dff Python-3.7.15 torch-1.12.1+cu113 CUDA:0 (Tesla T4, 15110MiB) Fusing layers... YOLOv5s summary: 213 layers, 7225885 parameters, 0 gradients WARNING ⚠ Environment does not support cv2.imshow() or PIL Image.show() [ WARN:0@5.409] global /io/opencv/modules/videoio/src/cap_v4l.cpp (902) open VIDEOIO(V4L2:/dev/video0): can't open camera by index Traceback (most recent call last): File "detect.py", line 258, in main(opt) File "detect.py", line 253, in main run(**vars(opt)) File "/usr/local/lib/python3.7/dist-packages/torch/autograd/grad_mode.py", line 27, in decorate_context return func(*args, **kwargs) File "detect.py", line 103, in run dataset = LoadStreams(source, img_size=imgsz, stride=stride, auto=pt, vid_stride=vid_stride) File "/content/yolov5/utils/dataloaders.py", line 364, in __init__ assert cap.isOpened(), f'{st}Failed to open {s}' AssertionError: 1/1: 0... Failed to open 0

I ran it and came across this error. Your suggestions please?

You should be able to run it in colab, but I believe you are detting an error because your webcam is not going to be connected to the instance. You will need to run it on a video file.

@@robmulla thanks for the response. Yeah, it is accessing the webcam through colab. Is there any way to connect to the webcam?

not*

Are you sure you are referencing the correct file/location?

@@robmulla yep ,That what i Found , raise NotImplementedError(f'ERROR: {w} is not a supported format') NotImplementedError: ERROR: yolov5x is not a supported format

Oh cool! I'm not sure how those videos were created but this would be really hard to pull off without a lot of false positives. You would essentially just need to use YOLO trained on a custom dataset with shoplifting labeled. I'm assuming it's also identifying the people from the pretrained yolo weights... check out my other video on how to train a custom detector.

@@robmulla Thanks much!

don’t use your camera unless you have to , i get issues too, switched to just detect the objects on my screen instead , still using the camera just doesn’t detect through the camera lol detects off of my screen

lol! You are correct! I love to YOLO and BOGO every day 😅 - this YOLO is different even though it has a similar name.

Ohh. That’s an interesting idea. Thanks for watching.

Thank you!

@@robmulla I do get some errors running this any ideas why? I’m on max if that helps

😅

Make sure you are using the correct web cam number for your system. Might be 0 or 1 and you change that in the command you run.

We shall see.

Oh really. I didn’t realize it changed. Check out my video with yolov7

It works for me. You might want to try yolov8 that was recently released.

@@robmulla ok But how can I use my GPU for running this. My cpu is utilized for this and the GPU is completely free. How to do this ....please help

Please reply