Why does the output processing (yolov7 tiny) takes so long compere to SSD_mobilenet ?
It makes the use of it not relevant...

why does it needs to go over 4 nested loops ?

openvino 2022.1版本不是有预处理的功能了吗，为什么还要用opencv来resize呢

./yolov7 yolov7.onnx horses.jpg 'CPU'

terminate called after throwing an instance of 'ov::Exception'
what(): vector::_M_range_check: __n (which is 1) >= this->size() (which is 1)
Aborted (core dumped)

From the ReadMe file link not working
7. Further optimization
Try this notebook (yolov7-optimization) and quantize your YOLOv7 model to INT8.

https://github.com/openvinotoolkit/openvino_notebooks/tree/develop/notebooks/225-yolov7-optimization

Hello, I followed this notebook to convert yolov7-e6 model to IR, then quantized the model on my own dataset with dataloader's image size 1280
https://github.com/openvinotoolkit/openvino_notebooks/blob/main/notebooks/226-yolov7-optimization/226-yolov7-optimization.ipynb
When I run inference with the model using main_preprocessing.cpp, I get weird results where the bbox does not line up with the object.

All I changed was int img_h = 1280 and int img_w = 1280; Do I need to change anything else? This didn't happen when I convert yolov7.pt model and quantize on the same dataset with image size 640

Do these directions in the repo need be updated??

I cant help but notice I dont think an export.py exists in the WongKinYiu/yolov7.git anymore??

  $ git clone [email protected]:WongKinYiu/yolov7.git
  $ cd yolov7/models
  $ python export.py --weights yolov7.pt

Also what were the directions on converting yolo7.pt to onnx? I cant remember how todo that at the moment. Not the tiny version but the regular larger size I could swear you used to be able to do that right in the WongKinYiu repo code.

Hi, thanks for providing this repo
I found a bug in yolov7.py.
It will cause producing incorrect outputs.

The code in line 228 should be img = img.transpose(2, 0, 1) rather than img.transpose(2, 0, 1).

NumPy's transpose operation does not support in-place assignment.
Reference: https://numpy.org/doc/1.23/reference/generated/numpy.ndarray.transpose.html

Hello @OpenVINO-dev-contest ,
thank you for this awesome work.
Does this script work with yolov7-seg? I have trained and exported my yolov7-seg to Onnx format. but I can't run it with this script, the following error appeared while using the python script:

Traceback (most recent call last):
  File "python/image.py", line 23, in <module>
    yolov7_detector.infer_image(args.input)
  File "/home/mad/YOLOv7_OpenVINO_cpp-python/python/yolov7.py", line 237, in infer_image
    self.infer_queue.wait_all()
  File "/home/mad/YOLOv7_OpenVINO_cpp-python/python/yolov7.py", line 186, in postprocess
    output.append(self.sigmoid(infer_request.get_output_tensor(0).data[batch_id].reshape(-1, self.size[0]*3, 5+self.class_num)))
ValueError: cannot reshape array of size 2948400 into shape (19200,85)

If possible any help or insight into this matter? I will appreciate it

Hey Ethan,

For the Python version....would it be really hard to modify the code to process multiple video feeds at once? Am thinking about tinkering around with this.

If it seems doable could you ever process and keep track in the pipe for detections in each feed, like hold some sort of memory for people detected in feed 1, feed 2, feed 3, and feed 4, etc... as frames are processed async style from all feeds?

Thanks!

Dump preprocessor: Input "images" (color BGR):
User's input tensor: [1,640,640,3], [N,H,W,C], f32
Model's expected tensor: [1,3,640,640], [N,C,H,W], f32
Pre-processing steps (2):
convert color (RGB): ([1,640,640,3], [N,H,W,C], f32, BGR) -> ([1,640,640,3], [N,H,W,C], f32, RGB)
scale (255,255,255): ([1,640,640,3], [N,H,W,C], f32, RGB) -> ([1,640,640,3], [N,H,W,C], f32, RGB)
Implicit pre-processing steps (1):
convert layout [N,C,H,W]: ([1,640,640,3], [N,H,W,C], f32, RGB) -> ([1,3,640,640], [N,C,H,W], f32, RGB)

thanks for repo. i was able to implement yolov7 repo on my system , its possible to implement it with --grid parameter? what changes i need to do? im getting size not match issue with --grid in your yolo.py script

What would I need to change to use 4 anchor boxes per output?

I am using the python script

is there any reference to adding deepsort tracker with .int8 or openvino ir format? should i need to convert deepsort weight to IR format or directly i need to use? i want to improve my fps speed on cpu

Would anyone be able to recommend any cheap hardware to run on even if its cheap CPU with poor frame rates? Am using it inside a building to count people in a conference room.

https://www.intel.com/content/www/us/en/products/details/nuc/kits/products.html

Or newest rasp berry pie with pie cam?

Hello,

Am testing out the Tiny Yolo in the IoT app I made which runs much faster than non Tiny yolo v7 on CPU. ~4 FPS running non tiny Yolo v7 and 8-10 with the Tiny Yolo and the asnyc inference.

The problem I have is some angles of people in conference room I get some poor object detection, like its very flaky. Would you have any suggestions to try?

Seems like lowering the confidence to self.conf_thres = 0.1 even still yields poor flaky classifying where the person in the screenshot is circled RED. I adjusted the parameter in the YOLOV7_OPENVINO class in yolo7.py:

class YOLOV7_OPENVINO(object):
    def __init__(self, model_path, device, pre_api, batchsize, nireq, use_flask):
        # set the hyperparameters
        self.classes = [
        "person", "bicycle", "car", "motorcycle", "airplane", "bus", "train", "truck", "boat", "traffic light",
        "fire hydrant", "stop sign", "parking meter", "bench", "bird", "cat", "dog", "horse", "sheep", "cow",
        "elephant", "bear", "zebra", "giraffe", "backpack", "umbrella", "handbag", "tie", "suitcase", "frisbee",
        "skis", "snowboard", "sports ball", "kite", "baseball bat", "baseball glove", "skateboard", "surfboard",
        "tennis racket", "bottle", "wine glass", "cup", "fork", "knife", "spoon", "bowl", "banana", "apple",
        "sandwich", "orange", "broccoli", "carrot", "hot dog", "pizza", "donut", "cake", "chair", "couch",
        "potted plant", "bed", "dining table", "toilet", "tv", "laptop", "mouse", "remote", "keyboard", "cell phone",
        "microwave", "oven", "toaster", "sink", "refrigerator", "book", "clock", "vase", "scissors", "teddy bear",
        "hair drier", "toothbrush"
       ]
        self.batchsize = batchsize
        self.use_flask = use_flask
        self.routes = webapp_utils.WebAppUtils()
        self.img_size = (640, 640) 
        self.conf_thres = 0.1

I'm still interested in quantization model compression on Yolo v7 (non-tiny) and compare the overall performance but having lots of issues trying to get the code to run....the notebook you linked in #5 , where when I run on Ubuntu the kernal dies half way and I cant get the code to run on Windows 10 but still trying.

Does this script work with yolov7-tiny?

Hi,

In the README is there a choice to do either Python or C++ on Ubuntu? Either do I need to do Python and the C++ step on Ubuntu?

Cool repo, thanks for making this...

I'm so sorry.Are you sure the cpp file is accurate?step8 is abnormal in my case

sometimes im getting 3 fps, then sometime its giving 10 fps,

when i print it in cmd its showing ~33 fps. i have applied simple logic for getting fps, because your code of fps is not working

fps=0
while:
                fps = (fps + (1. / (time.time() - t1))) / 2
                
                #fps=(stop_time-start_time)
                print("fps",fps)

if its possible can you add fps code & saving the resulting video? it would be really great for doing benchmark for fps

Does this repo support - YOLOv7 with Multiple Object Tracker - SORT Algorithm?

Or this I think:
https://youtu.be/keXpp8FhORI

Sort of curious if I can apply this to people walking through doorways inside a building : )

once it finishes running, I'm getting this error

im loading video in it

  File "python/webcam.py", line 25, in <module>
    yolov7_detector.infer_cam(args.input)
  File "D:\openVINO\YOLOv7_OpenVINO_grid\python\yolov7.py", line 295, in infer_cam
    img = self.letterbox(frame, self.img_size)
  File "D:\openVINO\YOLOv7_OpenVINO_grid\python\yolov7.py", line 60, in letterbox
    shape = img.shape[:2]  # current shape [height, width]
AttributeError: 'NoneType' object has no attribute 'shape'

hi,
I met a error of "非常量的表达式", when I build the c++ code with vs2019 .
I changed it to below to fix it.
"
float data1[6406403];
"

Hi would you have any tips for getting up and running on webcam.py? Am running this on Windows 10 with the Pypi version of openVino installed...

If I comment out the imshow in yolo7.py the code seems to run:

    def draw(self, img, boxinfo):
        for xyxy, conf, cls in boxinfo:
            self.plot_one_box(xyxy, img, label=self.classes[int(cls)], color=self.colors[int(cls)], line_thickness=2)
        #cv2.imshow('YOLOv7 results', img) 
        return 0

But doesn't appear to print the FPS when exiting (CNTRL-C in terminal) which I was hoping monitor frame rates with the code inside the infer_cam function:

            if c==27: 
                self.infer_queue.wait_all()
                break 
        cap.release() 
        cv2.destroyAllWindows() 
        end_time = time.time()
        # Calculate the average FPS\n",
        fps = count / (end_time - start_time)
        print("throughput: {:.2f} fps".format(fps))