How I can save the numpy value of output features as .csv format file layer by layer (stage by stage)?

[PariaDarbani]

How I can save the numpy value of output features as .csv format file layer by layer (stage by stage)?

I can access the weights of each layer by model.state_dict() but how I can print or save the input/output feature maps from selective layers?

Additional context

[glenn-jocher]

@PariaDarbani you can use the --visualize flag to output YOLOv5 feature maps:

python detect.py --visualize

See #3920

[PariaDarbani]

Thank you. it saves the images of outputs. how I can save the numpy values or tensor of each image (numbers)

[glenn-jocher]

@PariaDarbani there's no method to save the output values directly, but you can customize the code to return values here:

yolov5/models/yolo.py

Lines 158 to 160 in aa18599

	if visualize:
	feature_visualization(x, m.type, m.i, save_dir=visualize)

[PariaDarbani]

@PariaDarbani there's no method to save the output values directly, but you can customize the code to return values here:

yolov5/models/yolo.py

Lines 158 to 160 in aa18599

	if visualize:
	feature_visualization(x, m.type, m.i, save_dir=visualize)

Thank you so much. solved. :)