Update README.md
Browse files
README.md
CHANGED
|
@@ -184,11 +184,11 @@ Sampled VEDB frames were first processed using a deterministic common pipeline:
|
|
| 184 |
For the **Periph-NF condition**, the following condition-specific transformation was then applied before the shared SimCLR augmentations:
|
| 185 |
|
| 186 |
1. Obtain the participant's synchronized per-frame gaze location. When a reliable gaze location was unavailable, use the image center as the fallback fixation location.
|
| 187 |
-
2.
|
| 188 |
-
3.
|
| 189 |
-
4.
|
| 190 |
-
5.
|
| 191 |
-
6.
|
| 192 |
|
| 193 |
This condition-specific construction occurred **before** the shared SimCLR augmentation pipeline.
|
| 194 |
|
|
|
|
| 184 |
For the **Periph-NF condition**, the following condition-specific transformation was then applied before the shared SimCLR augmentations:
|
| 185 |
|
| 186 |
1. Obtain the participant's synchronized per-frame gaze location. When a reliable gaze location was unavailable, use the image center as the fallback fixation location.
|
| 187 |
+
2. Center a circular scotoma on the gaze-defined fixation location and fill the masked central region with uniform gray (`128`).
|
| 188 |
+
3. Gaussian-blur the scotoma boundary (`kernel = 15`) to produce a feathered transition between the masked central region and the retained peripheral image.
|
| 189 |
+
4. Apply [NeuroFovea](https://github.com/ArturoDeza/NeuroFovea_PyTorch) to the scotoma-masked frame, with the transformation centered on the gaze-defined fixation location.
|
| 190 |
+
5. Use the NeuroFovea scale parameter **`s = 0.4`**, producing an eccentricity-dependent, metamer-like transformation in which precise spatial structure is progressively replaced by texture-like information farther from fixation.
|
| 191 |
+
6. Resize the resulting NeuroFovea-transformed image to the network's **`224 × 224`** input resolution, if necessary.
|
| 192 |
|
| 193 |
This condition-specific construction occurred **before** the shared SimCLR augmentation pipeline.
|
| 194 |
|