DM-Diaz commited on
Commit
89cf408
Β·
verified Β·
1 Parent(s): 41a2caf

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +28 -8
README.md CHANGED
@@ -21,7 +21,7 @@ tags:
21
 
22
  This repository contains **subject-specific NSD voxelwise encoding-model fits and the variance-partitioning fits reported in the associated study** from:
23
 
24
- **Diaz, D. M., & Henderson, M. M. (2026). _Eccentricity-Constrained CNN Training Reveals Adaptive Information Coding Around the Visual Field._ Proceedings of the 9th Conference on Cognitive Computational Neuroscience.**
25
 
26
  **DOI:** [10.32470/0416gfsq](https://doi.org/10.32470/0416gfsq)<br>
27
  **arXiv:** [2607.19316](https://arxiv.org/abs/2607.19316)<br>
@@ -67,7 +67,8 @@ The four pretrained models were trained on complementary versions of the same so
67
 
68
  | Component | Status |
69
  | --- | --- |
70
- | Subject-specific encoding-model fits | Available |
 
71
  | Variance-partitioning fits reported in the paper | Available |
72
  | Model card | Available |
73
  | Feature-extraction / model-fitting code | Forthcoming |
@@ -84,14 +85,18 @@ baseline/
84
  fovea-gaze/
85
  periph/
86
  periph-nf/
 
 
 
 
87
  variance-partitioning/
88
  β”œβ”€β”€ fovea-gaze-vs-periph/
89
  └── periph-vs-periph-nf/
90
  ```
91
 
92
- Each model-condition folder contains voxelwise encoding-model fits for **NSD subjects S1–S8**. The `variance-partitioning/` directory contains the two model comparisons reported in Figure 4C–D of the paper, with one fit for each of the same eight NSD participants.
93
 
94
- Example:
95
 
96
  ```text
97
  fovea-gaze/
@@ -99,6 +104,14 @@ fovea-gaze/
99
  β”œβ”€β”€ NSD_S2_resnet18-Fovea-Gaze_concat.npy
100
  β”œβ”€β”€ ...
101
  └── NSD_S8_resnet18-Fovea-Gaze_concat.npy
 
 
 
 
 
 
 
 
102
  ```
103
 
104
  ## Architecture
@@ -117,13 +130,13 @@ Voxelwise encoding models were fit using fMRI responses from the [**Natural Scen
117
 
118
  For each participant, the **1,000 images shared across all NSD participants** were reserved as the final held-out evaluation set. The remaining approximately **9,000 participant-specific images** were used for model fitting and regularization selection. Within this fitting set, a nested holdout partition was used to select the ridge penalty independently for each voxel.
119
 
120
- The same intact `224 Γ— 224` NSD images were presented to each of the four pretrained ResNet-18 models. The Baseline, Fovea-Gaze, Periph, and Periph-NF transformations were applied only during VEDB SimCLR pretraining and were **not reapplied to NSD stimuli** during feature extraction.
121
 
122
  Encoding performance was evaluated on the held-out 1,000-image set using voxelwise coefficient of determination (`RΒ²`) and Pearson correlation (`corr`). These held-out metrics are stored directly in each released encoding-model fit.
123
 
124
  ## Encoding-Model Procedure
125
 
126
- The encoding models were fit separately for each **NSD subject (S1–S8)** and each VEDB-pretrained model condition.
127
 
128
  NSD stimuli were presented to the pretrained ResNet-18 models as intact `224 Γ— 224` images. The Baseline, Fovea-Gaze, Periph, and Periph-NF transformations were used during SimCLR pretraining and were **not reapplied to NSD images** during encoding-model evaluation.
129
 
@@ -166,7 +179,7 @@ Performance on the final held-out set was quantified with voxelwise coefficient
166
  Each `.npy` file contains a Python dictionary with the fitted encoding model and associated metadata:
167
 
168
  - `subject` β€” NSD subject number
169
- - `model` β€” visual model condition
170
  - `features_file_list` β€” feature files used for model fitting
171
  - `lambdas` β€” candidate ridge penalties
172
  - `weights` β€” fitted voxelwise ridge-regression weights
@@ -190,6 +203,11 @@ The released `.npy` files retain the internal model identifiers used during the
190
  | Fovea-Gaze | `resnet18-FoveaGaze` |
191
  | Periph | `resnet18-PeriphNonTTM` |
192
  | Periph-NF | `resnet18-PeriphTTM` |
 
 
 
 
 
193
 
194
  These internal identifiers are preserved in fields such as `model`, `model1`, `model2`, and the keys of the variance-partitioning dictionaries. They reflect the naming convention used during model development and do not indicate additional model conditions.
195
 
@@ -304,7 +322,9 @@ voxel_nc = fit["voxel_nc"]
304
 
305
  ## Scope of the Released Artifacts
306
 
307
- The repository contains **32 subject-specific encoding-model fits** (4 model conditions Γ— 8 NSD participants) and **16 variance-partitioning fits** corresponding to the two variance-partition analyses reported in the paper (2 model comparisons Γ— 8 participants), for a total of **48 fitted `.npy` artifacts**.
 
 
308
 
309
  The released files contain fitted voxelwise weights, held-out prediction metrics, ridge-penalty information, and voxel metadata used in the study. The variance-partitioning artifacts additionally contain the model-A-only, model-B-only, and combined fits used to compute unique variance.
310
 
 
21
 
22
  This repository contains **subject-specific NSD voxelwise encoding-model fits and the variance-partitioning fits reported in the associated study** from:
23
 
24
+ **Diaz, D. M., & Henderson, M. M. (2026). *Eccentricity-Constrained CNN Training Reveals Adaptive Information Coding Around the Visual Field.* Proceedings of the 9th Conference on Cognitive Computational Neuroscience.**
25
 
26
  **DOI:** [10.32470/0416gfsq](https://doi.org/10.32470/0416gfsq)<br>
27
  **arXiv:** [2607.19316](https://arxiv.org/abs/2607.19316)<br>
 
67
 
68
  | Component | Status |
69
  | --- | --- |
70
+ | VEDB subject-specific encoding-model fits | Available |
71
+ | Reference-model encoding fits (STL-10, ImageNet-100, ImageNet-1K) | Available |
72
  | Variance-partitioning fits reported in the paper | Available |
73
  | Model card | Available |
74
  | Feature-extraction / model-fitting code | Forthcoming |
 
85
  fovea-gaze/
86
  periph/
87
  periph-nf/
88
+ reference-models/
89
+ β”œβ”€β”€ stl10/
90
+ β”œβ”€β”€ imagenet-100/
91
+ └── imagenet-1k/
92
  variance-partitioning/
93
  β”œβ”€β”€ fovea-gaze-vs-periph/
94
  └── periph-vs-periph-nf/
95
  ```
96
 
97
+ Each encoding-model folder contains voxelwise fits for **NSD subjects S1–S8**. The four VEDB folders contain fits derived from the Baseline, Fovea-Gaze, Periph, and Periph-NF models. The `reference-models/` directory contains the corresponding fits derived from the STL-10, ImageNet-100, and ImageNet-1K reference models. The `variance-partitioning/` directory contains the two model comparisons reported in Figure 4C–D of the paper, with one fit for each of the same eight NSD participants.
98
 
99
+ Examples:
100
 
101
  ```text
102
  fovea-gaze/
 
104
  β”œβ”€β”€ NSD_S2_resnet18-Fovea-Gaze_concat.npy
105
  β”œβ”€β”€ ...
106
  └── NSD_S8_resnet18-Fovea-Gaze_concat.npy
107
+
108
+ reference-models/
109
+ β”œβ”€β”€ stl10/
110
+ β”‚ └── NSD_S1_resnet18-simclr-stl10_concat.npy
111
+ β”œβ”€β”€ imagenet-100/
112
+ β”‚ └── NSD_S1_resnet18-simclr-imagenet100_concat.npy
113
+ └── imagenet-1k/
114
+ └── NSD_S1_resnet18-simclr-imagenet1k_concat.npy
115
  ```
116
 
117
  ## Architecture
 
130
 
131
  For each participant, the **1,000 images shared across all NSD participants** were reserved as the final held-out evaluation set. The remaining approximately **9,000 participant-specific images** were used for model fitting and regularization selection. Within this fitting set, a nested holdout partition was used to select the ridge penalty independently for each voxel.
132
 
133
+ The same intact `224 Γ— 224` NSD images were presented to each of the seven pretrained ResNet-18 models. The Baseline, Fovea-Gaze, Periph, and Periph-NF transformations were applied only during VEDB SimCLR pretraining and were **not reapplied to NSD stimuli** during feature extraction. The same NSD feature-extraction procedure was used for the STL-10, ImageNet-100, and ImageNet-1K reference models.
134
 
135
  Encoding performance was evaluated on the held-out 1,000-image set using voxelwise coefficient of determination (`RΒ²`) and Pearson correlation (`corr`). These held-out metrics are stored directly in each released encoding-model fit.
136
 
137
  ## Encoding-Model Procedure
138
 
139
+ The encoding models were fit separately for each **NSD subject (S1–S8)** and each of the seven pretrained visual models.
140
 
141
  NSD stimuli were presented to the pretrained ResNet-18 models as intact `224 Γ— 224` images. The Baseline, Fovea-Gaze, Periph, and Periph-NF transformations were used during SimCLR pretraining and were **not reapplied to NSD images** during encoding-model evaluation.
142
 
 
179
  Each `.npy` file contains a Python dictionary with the fitted encoding model and associated metadata:
180
 
181
  - `subject` β€” NSD subject number
182
+ - `model` β€” visual model identifier
183
  - `features_file_list` β€” feature files used for model fitting
184
  - `lambdas` β€” candidate ridge penalties
185
  - `weights` β€” fitted voxelwise ridge-regression weights
 
203
  | Fovea-Gaze | `resnet18-FoveaGaze` |
204
  | Periph | `resnet18-PeriphNonTTM` |
205
  | Periph-NF | `resnet18-PeriphTTM` |
206
+ | STL-10 | `resnet18-pretrained-simclr` |
207
+ | ImageNet-100 | `resnet18-simclr-imgnet100` |
208
+ | ImageNet-1K | `resnet18-simclr-imgnet1k` |
209
+
210
+ The internal identifier `resnet18-pretrained-simclr` corresponds to the STL-10 SimCLR ResNet-18 checkpoint obtained from the external [Spijkervet/SimCLR](https://github.com/Spijkervet/SimCLR) release.
211
 
212
  These internal identifiers are preserved in fields such as `model`, `model1`, `model2`, and the keys of the variance-partitioning dictionaries. They reflect the naming convention used during model development and do not indicate additional model conditions.
213
 
 
322
 
323
  ## Scope of the Released Artifacts
324
 
325
+ The repository contains **56 subject-specific encoding-model fits** across seven pretrained visual models: **32 fits from the four VEDB-pretrained models** (4 models Γ— 8 NSD participants) and **24 fits from the three non-egocentric reference models** (3 models Γ— 8 NSD participants).
326
+
327
+ The repository additionally contains **16 variance-partitioning fits** corresponding to the two variance-partition analyses reported in the paper (2 model comparisons Γ— 8 participants), for a total of **72 fitted `.npy` artifacts**.
328
 
329
  The released files contain fitted voxelwise weights, held-out prediction metrics, ridge-penalty information, and voxel metadata used in the study. The variance-partitioning artifacts additionally contain the model-A-only, model-B-only, and combined fits used to compute unique variance.
330