RuntimeError: shape '[512, 512, 3]' is invalid for input of size 32768

[yassinesefrioui]

Hi !
in my file configuration (yolov4.cfg), i'm using this parameters :

width=512
height=512
channels=1

in cfg.py (YOLOv4_pytorch/tool/cfg.py), i'm using this parameters :

    prev_width = 512
    prev_height = 512
    prev_filters = 1
```

i get all time this error : 
  File "./demo.py", line 43, in <module>
    detect(cfgfile, weightfile, imgfile)
  File "./demo.py", line 28, in detect
    boxes = do_detect(m, sized, 0.5, 0.4, use_cuda)
  File "/home/yassines/Downloads/YOLOv4_Pytorch/utils/utils.py", line 357, in do_detect
    img = img.view(height, width, 3).transpose(0, 1).transpose(0, 2).contiguous()
RuntimeError: shape '[512, 512, 3]' is invalid for input of size 32768


i think, i need to change something in this code : 
```
        img = img.view(height, width, 3).transpose(0, 1).transpose(0, 2).contiguous()
        # batch:1  channel:3
        img = img.view(1, 3, height, width)
        # 除以255，归一化
```
how i can change it to take just One channel not three channel ? 

thank you 

[bobo0810]

Hi !
in my file configuration (yolov4.cfg), i'm using this parameters :

width=512
height=512
channels=1

in cfg.py (YOLOv4_pytorch/tool/cfg.py), i'm using this parameters :

    prev_width = 512
    prev_height = 512
    prev_filters = 1

i get all time this error :
File "./demo.py", line 43, in
detect(cfgfile, weightfile, imgfile)
File "./demo.py", line 28, in detect
boxes = do_detect(m, sized, 0.5, 0.4, use_cuda)
File "/home/yassines/Downloads/YOLOv4_Pytorch/utils/utils.py", line 357, in do_detect
img = img.view(height, width, 3).transpose(0, 1).transpose(0, 2).contiguous()
RuntimeError: shape '[512, 512, 3]' is invalid for input of size 32768

i think, i need to change something in this code :

        img = img.view(height, width, 3).transpose(0, 1).transpose(0, 2).contiguous()
        # batch:1  channel:3
        img = img.view(1, 3, height, width)
        # 除以255，归一化

how i can change it to take just One channel not three channel ?

thank you

Hello i am sorry just noticed...
Do you mean you want to use a grayscale image with channel 1 for testing? Since the code is not written by myself, and the pre-training model is based on color images, I guess the effect of testing grayscale images is not good.