search

hitachinsk / FGT

[ECCV 2022] Flow-Guided Transformer for Video Inpainting
https://hitachinsk.github.io/publication/2022-10-01-Flow-Guided-Transformer-for-Video-Inpainting
MIT License
311 stars 32 forks source link

FGT model config problem #16

Closed OHaiYo-lzy closed 2 years ago

OHaiYo-lzy commented 2 years ago

When I tried to train your FGT model ,it occured to me that I`m missing the right .yaml config file in flowCheckPoint dir which caused this to happen

RuntimeError: Error(s) in loading state_dict for Model: Missing key(s) in state_dict:

.. Could pls release the right and corresponding config file?

hitachinsk commented 2 years ago

I have appended the config (yaml) file in the pretrained models. Please redownload the weight.zip as instructed in the README. I also update the corresponding codes, therefore you also need to update the codebase. Sorry for the late reply.

OHaiYo-lzy commented 2 years ago

Thx for your reply. But after I tried to do so , the same problem occured that firstly: config['resBlocks'], config['use_bias'], config['conv_type'], config['multi_out'], KeyError: 'multi_out'

so I add multi_out: 0 into the config.yaml, then: RuntimeError: Error(s) in loading state_dict for Model: Missing key(s) in state_dict: "net.edgeDetector.projection.featureConv.weight", "net.edgeDetector.projection.featureConv.bias", "net.edgeDetector.mid_layer_1.featureConv.weight", "net.edgeDetector.mid_layer_1.featureConv.bias", "net.edgeDetector.mid_layer_2.featureConv.weight", "net.edgeDetector.mid_layer_2.featureConv.bias", "net.edgeDetector.out_layer.featureConv.weight", "net.edgeDetector.out_layer.featureConv.bias".

or multi_out: 1: RuntimeError: Error(s) in loading state_dict for Model: Missing key(s) in state_dict: "net.out4.featureConv.weight", "net.out4.featureConv.bias", "net.out2.featureConv.weight", "net.out2.featureConv.bias", "net.edgeDetector.projection.featureConv.weight", "net.edgeDetector.projection.featureConv.bias", "net.edgeDetector.mid_layer_1.featureConv.weight", "net.edgeDetector.mid_layer_1.featureConv.bias", "net.edgeDetector.mid_layer_2.featureConv.weight", "net.edgeDetector.mid_layer_2.featureConv.bias", "net.edgeDetector.out_layer.featureConv.weight", "net.edgeDetector.out_layer.featureConv.bias".

Could pls look into that and help to fix that? Thx!!

OHaiYo-lzy commented 2 years ago

By the way ,I noticed that you updated the flow_net model file to lafc_single file , But didn`t update the corresponding network file.Hopefully be handled,Thx

hitachinsk commented 2 years ago

I checked the details, and found the problem was caused by the wrong model I uploaded. I have uploaded the right weights and modified the corresponding codes, please download this project again. I tested the codes on my own server, and the training process could be executed. If you have any problems, please reply under this comment. Thanks for pointing out my mistake.