Upload 4 files

README.md

@@ -1,3 +1,32 @@
----
-license: mit
----
+# Pretrained Checkpoints
+
+| File              | Train source | Paper Table 2 mIoU | Paper Table 3 mIoU |
+|-------------------|--------------|-------------------:|-------------------:|
+| `bocchi.pth`      | BOCCHI       |              0.850 |              0.403 |
+| `reloblur.pth`    | ReLoBlur     |              0.930 |              0.388 |
+| `blurdataset.pth` | BlurDataset  |              0.640 |              0.276 |
+
+All checkpoints are **MultiTransformUNet** (alias `DCTEverywhereUNet`,
+referred to in the paper as **MSDCT-UNet**): same architecture, different
+training source. Each is ~117 MB (FP32).
+
+## Download
+
+> **Anonymous review note**: pretrained weights are hosted on an external
+> service that does not reveal author identity. Replace the URLs below with
+> the de-anonymised HuggingFace links during the camera-ready phase.
+
+```bash
+# placeholder (to be replaced before camera-ready)
+mkdir -p checkpoints
+curl -L <ANONYMIZED_URL>/bocchi.pth      -o checkpoints/bocchi.pth
+curl -L <ANONYMIZED_URL>/reloblur.pth    -o checkpoints/reloblur.pth
+curl -L <ANONYMIZED_URL>/blurdataset.pth -o checkpoints/blurdataset.pth
+```
+
+## Verifying
+
+```bash
+python eval.py --config configs/bocchi.yaml --ckpt checkpoints/bocchi.pth --split val
+# Expected: mean_iou ~ 0.850
+```

blurdataset.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:365d9035f176f1e8684c0cbcde6deb42f4870b637e5d0533fbb99a5840a22200
+size 117885519

bocchi.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a7192e30e782b38408d96d38edd699ddf68d04c2c19ea4d66370d0c7b1ce037d
+size 117878181

reloblur.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4e27e176df028cff48fa657823726cfb7dba4579c2123140ea8c45cc69d604fe
+size 117884649