Error in loading pickle weights

bowenc0221 / panoptic-deeplab

This is Pytorch re-implementation of our CVPR 2020 paper "Panoptic-DeepLab: A Simple, Strong, and Fast Baseline for Bottom-Up Panoptic Segmentation" (https://arxiv.org/abs/1911.10194)

Apache License 2.0

585 stars 117 forks source link

Error in loading pickle weights #86

Closed 1201amit closed 3 years ago

1201amit commented 3 years ago

I am getting runtime error("Invalid magic number; corrupt file?") while loading pickle weights.

suyuzhang commented 3 years ago

I have the same problem. @bowenc0221 Can you provide a .pth pre-trained model of ResNet52 for cityscapes panoptic segmentation? Thanks!

1201amit commented 3 years ago

The issue is that panoptic-deeplab (this workspace) and detectron2 Panoptic-DeepLab have different key values in model parameters and given weights follow detectron2 keys. Best way is to create a model using detectron2 Panoptic-DeepLab.

bowenc0221 commented 3 years ago

@suyuzhang

The .pkl file for ResNet52 is converted from this code: https://github.com/facebookresearch/detectron2/blob/master/tools/convert-torchvision-to-d2.py

The model field is the corresponding .pth weights with some key conversion.