Official YOLO v7 Instance Segmentation COMPLETE TUTORIAL | Windows & Linux

[@sowmyaam7168]

I really love the way of your teaching, I will be waiting for your uploads. Please provide the video for how to do object tracking from scratch. It will be helpful for the learners.

[@TheCodingBug]

It will be uploaded by the end of this month.

[@sowmyaam7168]

@@TheCodingBug Thank you so much 🙂

[@davidlozano1927]

@@TheCodingBug Really happy to hear that video is coming up. I started to learn about this field and your videos are helping me a lot.

[@Laddu225]

Thank you very much for your video. Could you please tell me how to get constant color of the mask of the detected objects in a video?

[@TheCodingBug]

Instead and of random color generation for each detection in the detect function, define a fixed color for each class and then use that for each class id in detections.

[@Laddu225]

@@TheCodingBug Yeah, but since it is a instance segmentation, we'll have multiple instances belongs to one class. In that case, we need to have four constant colors of four persons without changing.

[@Laddu225]

@@TheCodingBug I tried it. Since, cars are belongs to one class, all cars are appearing in same color. But I need every instance of a object calls have a constant color through out the video. So, could you please help me to get it?

[@chukypedro818]

I am getting pred, and pred_masks as None, None. what could be the reasons for this?

[@willpayne9663]

Great video! I am running this on CPU only and it seems to throw an error of incompatible tensors, do you know how i can fix this? RuntimeError: Input type (torch.FloatTensor) and weight type (torch.HalfTensor) should be the same or input should be a MKLDNN tensor and weight is a dense tensor I need to change the code in the function processFrame(image) ?

[@mehmethalıcı010]

+1

[@melodieopale8345]

Hello, The issue is that the model must be cast into float type. in __main__ you have the following lines of code: if half: model = model.half() You must add the following lines: else: model = model.float() It should work after that.

[@dl2894]

Thanks for your video, it really help a lot. I am wondering what is the function of parameter "recompute_scle_factor...", and if I want to consistent the color for each instance object how can I do ? Appreciation for your willingness to share your thoughts.

[@AnshulParmar]

First of all thank you for such neat teaching... How to do detection for custom objects??

[@TheCodingBug]

I will make the tutorial in the next month on custom dataset.

[@yzx8828]

hello, thanks for your video, i have one question here, as detectron2 has already done the segmentation, why we still need yolov7 ? Does yolo v7 do the segmentation individually or it borrows the segmentation part from detectron2 ?

[@yeohweixiangbenjy1983]

Thank for the video. May I know is it possible to train custom 'mask.pt' file? I need it for my school work. Thx

[@QuarktaschemitSenf]

Do you offer training hours via Teams against payment?

[@TheCodingBug]

Not at the moment.

[@karthiksubramani8931]

Thanks for the Video but apart from Webcams, can we also detect from an external camera with an RTSP link? I would like to apply instance segmentation for an external camera (not Webcam) and would be really helpful if it would work!

[@TheCodingBug]

Yes. Just use rstp url as source for video instead of mp4 file.

[@imanbadr9283]

Thanks for the detailed and useful explanation. I am searching for a way to recognize all objects but then apply segmentation to hide persons. Is there a way to apply segmentation selectively using yolov7? Thanks in advance

[@dinanmutamaddin400]

i have been do same thing with ur tutorial but why only image showing, not show bonding box or segmentation over the image

[@abangfikri1865]

Thank u very much for the tutorial, but i have a question on if i can deploy this segmentation model on android? Is there a way for that to be done?

[@brunospfc8511]

Excelent content, please do Tracking / counting with this yolov7, thanks !

[@TheCodingBug]

Tracking tutorial is already made. I couldn't upload due to some health issues. Will post it by Tuesday.

[@atnguyen9841]

Thanks a lot for this. I have a question that how can I get prediction keypoints or coordinates of segmentation from the result ?

[@seul-kiyeom6222]

Nice video. Is there any way to do image classification based on yolov7 easily?

[@TheCodingBug]

Yes. I'll make a video next month.

[@seul-kiyeom6222]

@@TheCodingBug thank you very much!

[@rmas4955]

is the difference much to the yolov7?

[@dinanmutamaddin400]

u know why i failed isntall detectron2? that error with pycocotools

[@akhilsoni4109]

@TheCodingBug Thanks for the video! But, in line 34, when you call model on image, the output is a tuple instead of a dictionary and thus it gives error in line 36 when you compute inf_out, train_out, ... Can you please help?

[@巫鎔輿]

do you solve the question? I have same error, can tell me ? thank you

[@faizaanscnz]

would u help to to draw mask only body outline not fill the body with color

[@vitorvasconcelos.o8343]

Hey is it possible to train this implementation in a custom dataset? the same way we train yolov7 for object detection?

[@tugbayildizli3828]

Thank you so much. How do I train custom instance segmentation?

[@julianmeier5235]

As far as I know, the train.py in the mask-branch is still under development due to unresolved detectron2 dependencys. If you want to train on your custom dataset you should use the u7-Seg branch. It has been tested successfully and shows similar results to the mask-branch.

[@CrispyHands]

Thanks for the video. i got error NameError: name 'model' is not defined while make def processFrame(image): output = model(image) on line 34. any solution?

[@TheCodingBug]

Check spellings... It's not "model".

[@CrispyHands]

@@TheCodingBug what correct spelling then?

[@CrispyHands]

@@TheCodingBug cuz on your video, it said output = model(image)

[@김진수-e6l]

Hi sir, thankt for uploading great video. I have some errors when I load 'yolov7-mask.pt' I got these errors (torch.load('yolov7-mask.pt') -> AttributeError: Can't get attribute 'Merge' on ). But if I load 'yolo7-tiny.pt' error does not occur and the code runs very well. How can I load 'yolov7-mask.pt' like your videos?

[@TheCodingBug]

Make sure your version of torch is the same as mine.

[@moodiali7324]

@@TheCodingBug I am facing the same error although i installed the same torch and cuda version same like you (torch 1.11 and cuda 11.3), how to solve this ?

[@moodiali7324]

have you found a solution to this issue?

[@gemshunt9637]

Hi! How to reparameterize a model and convert to onnx?

[@michelangelo24]

Thanks a lot for this, man. ❤

[@dinanmutamaddin400]

how to training and deploy model

[@ahsanulhaque4811]

Hi, I am trying to do instance segementation with yolov7 using custom dataset, but I am struggling to convert coco style annotations to yolo style annotations. I know how to convert bounding boxes, but how to convert the segmentation list or RLE?

[@BlazenseiTV]

Hello sir have you already converted coco format annotation to Yolo style annotation? Because it's also my problem now, Thank You sir.

[@joart333]

I need to have gpu for instance segmentation ???

[@TheCodingBug]

If you want a decent FPS, then yes.

[@joart333]

but sir detectron will not install in cpu I think??

[@TheCodingBug]

@@joart333 I recall that it'll throw an error when installing pycocotools.... I will check that on the weekend and let you know.

[@joart333]

@@TheCodingBug OK sir

[@V7labs]

We want to collaborate! What is the best email?

[@justbendev2324]

0:18

[@amitprataprana322]

hello sir, while installing the detectron2 I get this error 5 errors detected in the compilation of "D:/Pothole_detection/detectron2/detectron2/layers/csrc/nms_rotated/nms_rotated_cuda.cu". nms_rotated_cuda.cu error: command 'C:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v11.6\\bin\ vcc.exe' failed with exit code 1

[@666zrock]

You should install ninja before detectron : conda install -c conda-forge ninja

[@DasKloputzer]

Great video, but why is it so slow? I thought Yolov7 makes up to 150 fps?

[@TheCodingBug]

That's on V100 GPU using YOLOV7 tiny.

[@DasKloputzer]

@@TheCodingBug ok thanks!

[@kishoreg8835]

i want yolov5 deepsparse custom training tutorial using collab. and convert .pt to .onnx and run .onnx for detection. if you do this perfectly i will give DONATE 1000rs. NO CAP!

[@jisi.techtooltips]

hi. I followed the instruction but why am I getting this error. Traceback (most recent call last): File "C:\Users\jisi\Desktop\YoloV7_Instance\yolov7-mask\segment.py", line 67, in onImage() NameError: name 'onImage' is not defined tia

[@TheCodingBug]

The method onImage have wrong spellings or you've missed something there