Parsing-R-CNN icon indicating copy to clipboard operation
Parsing-R-CNN copied to clipboard

Pretrained Resnext weights and demo

Open Naman-ntc opened this issue 4 years ago • 3 comments

Hi @soeaver, Thank you for releasing the excellent codebase. I wanted to ask if you would be releasing weights for your state of art model with ResNext101 backend (for CIHP or MHP datasets). I am also willing to train the ResNext backend model, could you please share the corresponding yaml config file with the recommended hyperparameters for that model.

Secondly, I wanted to ask if you have a demo code available which could be run on a custom dataset. Thanks,

Naman-ntc avatar May 30 '20 17:05 Naman-ntc

One thing you can do is to change the backbone in the yaml file and train the model from scratch. Also, you can copy some weights from the given pretrained imagenet checkpoint of ResNeXt-101.

BACKBONE:
  CONV_BODY: "resnext"

iamrishab avatar Aug 06 '20 13:08 iamrishab

@iamrishab Thank u for your reply! I wonder if it's convient for u to share the trained model using resnext101 as backbone for us to test and align ourselves' results. and @Naman-ntc have u trained and got the model?

ignitemylife avatar Aug 10 '20 09:08 ignitemylife

@iamrishab can you share your weight file on the google drive , thansk in adavance

abhigoku10 avatar Sep 13 '21 17:09 abhigoku10