Releases: bonlime/pytorch-tools
Self-Trained weights
Detection models weights
Pretrained weights for detection models.
EfficientDet models were trained by Google, then ported to PyTorch by @zylo117 in his repository and then mapped to models in this repo by @bonlime . Mapping gives absolute error < 1e-8
for raw outputs.
upd. additionally removed extra bias in first BiFPN layer downsample convs. Pretrained weights for them were ~1e-5
. So it doesn't affect validation results.
RetinaNet models are ported from mmdetection. Mmdetection ResNet is slightly different (stride 2 in conv1x1 instead of conv3x3) and order of anchors is also different so it's impossible to do inference using this weights but they work much better for transfer learning than starting from imagenet pretrain
Segmentation Models weights
Weights for segmentation models in this repo
Imagenet models weights
Releasing weights for models from
TResNet: High Performance GPU-Dedicated Architecture
: https://arxiv.org/pdf/2003.13630.pdf
Original code
: https://github.com/mrT23/TResNet
This weights were trained by authors and only ported to be loaded into models in this repo
Weights for HRNet were trained by authors as were uploaded to GitHub instead of OneDrive by me for easier loading.
Weights for Group Norm + Weight Standardization from original authors https://github.com/joe-siyuan-qiao/pytorch-classification/ . State dicts were remapped to allow easy loading into this repo models
Weights for Group Norm models R-50 & R-101 are from FAIR. They were originally trained in Caffe then ported to Detectron2 and then remapped by me to allow loading into this repo models.