lib/modeling · 9b41e84bd3b874b9749b27abcf0c448e6191bb92 · 人工智能 / detectron

Fix Retinanet training bug with num_classes != 81 · 1624f810

Moritz Kampelmuehler authored Feb 22, 2018

Summary:
Fixed bug preventing RetinaNet training with num_classes != 81 from converging.

Resolves #93
Closes https://github.com/facebookresearch/Detectron/pull/176

Reviewed By: ir413

Differential Revision: D7056134

Pulled By: rbgirshick

fbshipit-source-id: d3c62ce98a23a09191dd2720bc7189ee5490635d

1624f810

Name	Last commit	Last update
..
FPN.py		Loading commit data...
ResNet.py		Loading commit data...
VGG16.py		Loading commit data...
VGG_CNN_M_1024.py		Loading commit data...
__init__.py		Loading commit data...
detector.py		Loading commit data...
fast_rcnn_heads.py		Loading commit data...
generate_anchors.py		Loading commit data...
keypoint_rcnn_heads.py		Loading commit data...
mask_rcnn_heads.py		Loading commit data...
model_builder.py		Loading commit data...
name_compat.py		Loading commit data...
optimizer.py		Loading commit data...
retinanet_heads.py		Loading commit data...
rfcn_heads.py		Loading commit data...
rpn_heads.py		Loading commit data...