LibreDDRNetm-imagenet
Official DDRNet-23 ImageNet-1K classifier (top-1 error 24.1% as reported upstream), mirrored byte-for-byte
for LibreYOLO. LibreDDRNet uses it to initialize semantic-segmentation training, as the upstream recipe does.
Source
ydhongHIT/DDRNet at commit de0db317c5af4b0946231f475cf74ca4f51b8ac2.
The file DDRNet23_imagenet.pth is the official checkpoint linked from the upstream README
(Google Drive), unchanged: SHA-256 571959df93dce2db4b57d788362eb049a941e21b1b076a57112f8256216d1c08.
Copyright (c) 2020 xingkong. Licensed under the MIT License.
Modifications
None. LibreYOLO pins the SHA-256 above in libreyolo/models/ddrnet/convert.py and verifies every download.
At load time it keeps the tensors the classifier shares with the segmentation network and leaves DAPPM and
the segmentation heads at their random initialization, as upstream does.
Usage
A fresh model downloads this file when training starts:
from libreyolo import LibreDDRNet
model = LibreDDRNet(size="m")
model.train(data="cityscapes.yaml")
Or pass a local copy: LibreDDRNet(size="m", pretrained_backbone="DDRNet23_imagenet.pth").
License
MIT License, on an implied basis. The upstream repository links these weights from its README and states no separate terms for them, so LibreYOLO hosts them under the repository's MIT license on that implied basis. This is LibreYOLO's disclosed interpretation, not a license confirmed by the publisher. See the LICENSE and NOTICE files in this repository.
Dataset terms
These weights were trained on ImageNet-1K, whose terms of access are for non-commercial research and educational use.