Open tommiekerssies opened 1 year ago
Hi, sorry for delayed reply. For semantic segmentation, please refer to https://github.com/WXinlong/DenseCL/blob/main/benchmarks/detection/README.md. About object detection and semi classification, please refer to https://github.com/open-mmlab/mmselfsup/blob/openselfsup/docs/MODEL_ZOO.md
Thanks for your response. I notice for DenseCL they use contract_dilation=True. This is not possible in your DynamicResnet. Did you not use this in your experiments? Also, for the FCNHead, do you change it depending on the subnet? For example, do you scale the number of channels by 1/4 of the number of in_channels from the DynamicResnet?
I have manipulated the dynamic resnet in my mmsegmentation repo, it's really my fault to forget this, I have update the codes. And I didn't scale the number of channels of FCNHead.
Great, again, thanks a lot!
Hi, would you mind sharing the config for fine-tuning a subnet on Cityscapes? Thanks in advance!