Jiahao Cheng commited on
Commit ·
5fc652d
1
Parent(s): aa0ba11
docs: link HISTAI training dataset
Browse files- README.md +15 -1
- model_index.json +1 -1
README.md
CHANGED
|
@@ -20,6 +20,7 @@ tags:
|
|
| 20 |
- qwen2.5
|
| 21 |
datasets:
|
| 22 |
- david4real/HistGen
|
|
|
|
| 23 |
metrics:
|
| 24 |
- rouge
|
| 25 |
- meteor
|
|
@@ -34,7 +35,7 @@ model-index:
|
|
| 34 |
name: Case-level pathology synoptic report generation
|
| 35 |
dataset:
|
| 36 |
name: HISTAI case-report pairs
|
| 37 |
-
type: HISTAI
|
| 38 |
metrics:
|
| 39 |
- type: rouge
|
| 40 |
name: ROUGE-L
|
|
@@ -55,6 +56,7 @@ model-index:
|
|
| 55 |
[](https://arxiv.org/abs/2605.30716)
|
| 56 |
[](https://github.com/AtlasAnalyticsLab/PathoSynVLM)
|
| 57 |
[](https://atlasanalyticslab.github.io/PathoSynVLM/)
|
|
|
|
| 58 |
|
| 59 |
PathoSynVLM is a token-efficient vision-language model for generating
|
| 60 |
case-level pathology synoptic reports from one or more whole-slide
|
|
@@ -177,6 +179,18 @@ metadata preparation, Stage 1 alignment, Stage 2 case-level fine-tuning,
|
|
| 177 |
and evaluation. Machine-readable paper configurations and reported values
|
| 178 |
are maintained in the same code repository.
|
| 179 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 180 |
## Training Recipe
|
| 181 |
|
| 182 |
- Stage 1: train the two-layer MLP aligner on HistGen + REG2025 while
|
|
|
|
| 20 |
- qwen2.5
|
| 21 |
datasets:
|
| 22 |
- david4real/HistGen
|
| 23 |
+
- histai/HISTAI-metadata
|
| 24 |
metrics:
|
| 25 |
- rouge
|
| 26 |
- meteor
|
|
|
|
| 35 |
name: Case-level pathology synoptic report generation
|
| 36 |
dataset:
|
| 37 |
name: HISTAI case-report pairs
|
| 38 |
+
type: histai/HISTAI-metadata
|
| 39 |
metrics:
|
| 40 |
- type: rouge
|
| 41 |
name: ROUGE-L
|
|
|
|
| 56 |
[](https://arxiv.org/abs/2605.30716)
|
| 57 |
[](https://github.com/AtlasAnalyticsLab/PathoSynVLM)
|
| 58 |
[](https://atlasanalyticslab.github.io/PathoSynVLM/)
|
| 59 |
+
[](https://huggingface.co/datasets/histai/HISTAI-metadata)
|
| 60 |
|
| 61 |
PathoSynVLM is a token-efficient vision-language model for generating
|
| 62 |
case-level pathology synoptic reports from one or more whole-slide
|
|
|
|
| 179 |
and evaluation. Machine-readable paper configurations and reported values
|
| 180 |
are maintained in the same code repository.
|
| 181 |
|
| 182 |
+
## Training Data
|
| 183 |
+
|
| 184 |
+
The released Stage 2 checkpoint was fine-tuned on case-report pairs from
|
| 185 |
+
[HISTAI](https://huggingface.co/datasets/histai/HISTAI-metadata). The
|
| 186 |
+
official metadata repository is the starting point for dataset access and
|
| 187 |
+
links to the organ-specific WSI repositories used by HISTAI. See the
|
| 188 |
+
[HISTAI source documentation](https://github.com/HistAI/HISTAI) for the
|
| 189 |
+
dataset structure, subsets, citation, and access instructions.
|
| 190 |
+
|
| 191 |
+
HISTAI data remain subject to the dataset's CC BY-NC 4.0 license and
|
| 192 |
+
current access requirements.
|
| 193 |
+
|
| 194 |
## Training Recipe
|
| 195 |
|
| 196 |
- Stage 1: train the two-layer MLP aligner on HistGen + REG2025 while
|
model_index.json
CHANGED
|
@@ -8,7 +8,7 @@
|
|
| 8 |
},
|
| 9 |
"dataset": {
|
| 10 |
"name": "HISTAI case-report pairs",
|
| 11 |
-
"type": "HISTAI"
|
| 12 |
},
|
| 13 |
"metrics": [
|
| 14 |
{
|
|
|
|
| 8 |
},
|
| 9 |
"dataset": {
|
| 10 |
"name": "HISTAI case-report pairs",
|
| 11 |
+
"type": "histai/HISTAI-metadata"
|
| 12 |
},
|
| 13 |
"metrics": [
|
| 14 |
{
|