Add Faster R-CNN R50-FPN COCO weights for TTA harness
Browse files- README.md +18 -0
- model.safetensors +3 -0
README.md
ADDED
|
@@ -0,0 +1,18 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
tags:
|
| 4 |
+
- object-detection
|
| 5 |
+
- test-time-adaptation
|
| 6 |
+
- faster-rcnn
|
| 7 |
+
- coco
|
| 8 |
+
---
|
| 9 |
+
# TTA-COCO-FasterRCNN-R50
|
| 10 |
+
|
| 11 |
+
Faster R-CNN ResNet-50-FPN trained on COCO (91-class id space), re-hosted for the
|
| 12 |
+
TTA-Evaluation-Harness as the two-stage detection source model (sibling of
|
| 13 |
+
`WNJXYK/TTA-COCO-DETR-R50`).
|
| 14 |
+
|
| 15 |
+
`model.safetensors` is the torchvision `fasterrcnn_resnet50_fpn` COCO `state_dict`
|
| 16 |
+
(`FasterRCNN_ResNet50_FPN_Weights.DEFAULT`, Apache-2.0). The harness loads it into
|
| 17 |
+
a real-`BatchNorm2d` backbone (`weights_backbone=None`, `strict=False`) so BN-based
|
| 18 |
+
test-time adaptation has a genuine adaptation surface. Pinned by `revision: v1.0`.
|
model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:c264932235bbe0ad1e7ce300af9bd8c18738fb609853f1168902d71661e2e507
|
| 3 |
+
size 167476736
|