ercanhocaninyeri

The problem still remains @AyushExel. I tried 4 different cases: Case 1- when I run `model.val()`, the output is this: Speed: 0.3ms preprocess, 8.5ms inference, 0.0ms loss, 2.0ms postprocess per...

I understand that the command `torch.cuda.synchronize()` should be inserted before calling `time()` function. So I run the following code five times: ``` torch.cuda.synchronize() t0 = time.time() results = model_loaded.predict(source=im2) torch.cuda.synchronize()...

Yes I am warming up the model as you suggested. This is the reason why I put the screenshot of the spreadsheet. See the output of time() for Run #1....