Bug when training using Roboflow-Custom-Scaled-YOLOv4 colab

[robomims]

Screen shots of error:

2. Copy/Paste of your code
   I used the git copy at:
   !git clone GitHub - roboflow/ScaledYOLOv4

Failure is on training statement:
!python train.py --img 768 --batch 16 --epochs 60 --data {dataset.location}/data.yaml --cfg ./models/yolov4-csp.yaml --weights ‘’ --name yolov4-csp-results --cache

3. What you’ve tried so far to debug
   I’ve run this Colab as recently as July, and was able to get it to work with a few mods to the utils/general.py code. Just trying to run same colab with different image training dataset.

This is the error message:
File “/content/ScaledYOLOv4/utils/general.py”, line 556, in build_targets
indices.append((b, a, gj.clamp_(0, gain[3]), gi.clamp_(0, gain[2]))) # image, anchor, grid indices
RuntimeError: result type Float can’t be cast to the desired output type long int

I haven’t tried to change the general.py code this time. It is different from the version that I was successfully using back in July.

[Mohamed]

@robomims Here is how to solve the issue, fully diagnosed:

Error: Roboflow-Custom-Scaled-YOLOv4

I found the Issue also reported in the YOLOv7 repo (creator of Scaled-YOLOv4 also made YOLOv7). It seems they introduced the same error in YOLOv7. This comment notes how to fix it: RuntimeError: result type Float can't be cast to the desired output type long int · Issue #35 · WongKinYiu/yolov7 · GitHub The fix is to convert to long the "gain" of the gridspace on declaration. In all lines with gain = torch.ones(7, device=targets.device) change to gain = torch.ones(7, device=targets.device).…

Error: Roboflow-Custom-Scaled-YOLOv4

Updated ScaledYOLOv4/utils/general.py (Google Drive download) Once this PR is merged to the yolov4-large branch of the Scaled-YOLOv4 repo fork, you’ll be able to run the notebook while avoiding making any of the manual changes noted above ^ : update line 512 and line 1101 of utils/general.py by mo-traor3-ai · Pull Request #2 · roboflow/ScaledYOLOv4 · GitHub

[Marlene]

@Mohamed , the fix works ! Thanks so much