christian

Thank you for your reply. I think yes. I set this: ignore_logdet=True.

I don't know why when I use `gbif_dl.io.download`, it returns: AttributeError: module 'gbif_dl' has no attribute 'io' How can I solve this? Thank you so much

Hi, thank you so much for your help, but there's another error occurs. I only download 1000 images, but it shows no space, which shouldn't be like that. download(data_generator,root="/export/data/yliu/dataset",random_subsets={'train': 0.95,...

You have to make some adaption on the ultralytics model to make the model truely pruned, otherwise the pruned model can not be fine-tuned, here's my implementation, https://github.com/Christian-lyc/model_compression-citizen-science-/blob/main/str_prune/stru_prune.py

Thank you so much for the quick reply. Actually, I want to use structure pruning and remove the zero-out channels. But Ultralytics model is written on yaml. So I tried...

Yes, I can see, but in structure pruning, if I pruning channels in one layer, the input channels of consecutive layer also need to be changed. I don't think I...

Yes, thank you. Do you also provide pytorch model code that can load the model ultralytics yolov8n pretrained model? Thank you

> > I clearly see the model become smaller > > How did you see it? From the module structure, I can see it changes in pycharm debug mode

> Is the speed faster? No,because the saved onnx model has same size with unpruned model, I don’t think it will be faster. I think the pruned channels become 0...

> What did you prune? `model.model`? > > You can try: > > ```python > model.ckpt.update(dict(model=model.model)) > del model.ckpt["ema"] > model.save("pruned.pt") > ``` > > And then loading `pruned.pt` and...