Remove internal checkpoint conversion note from model card
Browse files
README.md
CHANGED
|
@@ -30,21 +30,13 @@ built on [vSHARP](https://arxiv.org/abs/2309.09954) inside the
|
|
| 30 |
|
| 31 |
| File | Role |
|
| 32 |
|------|------|
|
| 33 |
-
| `uniform_vsharp.pt` |
|
| 34 |
| `uniform_brain.yaml` | Brain inference (default **4×** FastMRIRandom, ACS 0.08) |
|
| 35 |
| `uniform_knee.yaml` | Knee inference (default **4×** FastMRIEquispaced, ACS 0.08) |
|
| 36 |
| `uniform_prostate.yaml` | Prostate inference (default **4×** FastMRIEquispaced, ACS 0.08) |
|
| 37 |
| `uniform_cardiac.yaml` | Cardiac / CMRxRecon inference (default **4×** FastMRIEquispaced, ACS 0.08) |
|
| 38 |
| `uniform_overview.png` | Overview figure |
|
| 39 |
|
| 40 |
-
### Checkpoint note
|
| 41 |
-
|
| 42 |
-
`uniform_vsharp.pt` was converted from the original Kosmos training checkpoint
|
| 43 |
-
(`model_686000.pt`) to the ModConv U-Net key layout used by current DIRECT, **including**
|
| 44 |
-
denoiser `conv_out.bias` tensors. Load it with the shipped YAMLs
|
| 45 |
-
(`image_unet_conv_out_bias: true`). You do **not** need
|
| 46 |
-
`convert_uniform_checkpoint.py` for Hub weights.
|
| 47 |
-
|
| 48 |
## Install DIRECT
|
| 49 |
|
| 50 |
```bash
|
|
|
|
| 30 |
|
| 31 |
| File | Role |
|
| 32 |
|------|------|
|
| 33 |
+
| `uniform_vsharp.pt` | Pretrained weights — use with the YAMLs below |
|
| 34 |
| `uniform_brain.yaml` | Brain inference (default **4×** FastMRIRandom, ACS 0.08) |
|
| 35 |
| `uniform_knee.yaml` | Knee inference (default **4×** FastMRIEquispaced, ACS 0.08) |
|
| 36 |
| `uniform_prostate.yaml` | Prostate inference (default **4×** FastMRIEquispaced, ACS 0.08) |
|
| 37 |
| `uniform_cardiac.yaml` | Cardiac / CMRxRecon inference (default **4×** FastMRIEquispaced, ACS 0.08) |
|
| 38 |
| `uniform_overview.png` | Overview figure |
|
| 39 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 40 |
## Install DIRECT
|
| 41 |
|
| 42 |
```bash
|