Updated README.md with configs and dataset details

README.md

@@ -1,15 +1,28 @@
 ---
 pretty_name: LeafNet
 dataset_info:
-  - config_name: train
-    splits:
-      - name: train
-        num_examples: 121337
+- config_name: train
+  splits:
+  - name: train
+    num_examples: 121337
 configs:
-  - config_name: train
-    data_files:
-      - split: train
-        path: "train-*.parquet"
+- config_name: train
+  data_files:
+  - split: train
+    path: train-*.parquet
+license: cc-by-4.0
+task_categories:
+- image-text-to-text
+- visual-question-answering
+language:
+- en
+tags:
+- vqa
+- agriculture
+- vision-language
+- computer-vision
+size_categories:
+- 100K<n<1M
 ---
 
 # LeafNet
@@ -29,14 +42,13 @@ to 72% accuracy, while the domain fine-tuned SCOLD model reached 99.15% on
 disease identification.
 
 ## Notes
-**This repo: 121,337 rows / 121,337 unique images (8.6GB)** — the public
-~70% subset of the full 186,000-image dataset (the remainder is held out
+The public dataset is just ~70% subset of the full 186,000-image dataset (the remainder is held out
 by the original authors).
 
 ## Layout
 
 Single config (`train`), matching every other AgML-standardized dataset's
-storage format — metadata parquet at the repo root, image shards under
+storage format with metadata parquet at the repo root, image shards under
 `images/`:
 
 ```
@@ -50,24 +62,14 @@ LeafNet_P/
 ```
 
 `images` holds `{"bytes": None, "path": ...}` per row's single image.
-Unlike MIRAGE, LeafNet's source data already carries a real, globally
-unique filename per image (e.g. `000000_Healthy_747.jpg`), so — as with
-AgroOmni — that original path is used directly as the shard-lookup key,
-with no content hashing needed.
-
-`raw_metadata` keeps the original (messy, Windows-style) source
-`file_name` path for provenance, the verbatim `caption`, and fields parsed
-from the caption template: `species`, `condition` (disease name, or
-`"Healthy"`), `symptoms`, and `healthy` (bool).
-
+`raw_metadata` keeps the original source `file_name` path for provenance, the verbatim `caption`.
 `messages` is a single-turn conversion, since the source data is an
 image-captioning dataset (one caption per image, no original question
-field): a fixed instruction prompt asking the model to describe the leaf's
-condition, with the original caption as the assistant's answer.
-
-**Not auto-loadable via `load_dataset()`.** Requires the companion loader
-(`load_agml_dataset()`), which reads images lazily and directly out of the
-zip shards — no extraction step.
+field): a fixed instruction prompt (mentioned below) asking the model to describe the leaf's
+condition, with the original caption as the assistant's answer was used to convert it to conversation.
+```
+USER_PROMPT = "Describe the condition of this plant leaf, including any visible disease and its symptoms."
+```
 
 ## Usage
 
@@ -111,4 +113,4 @@ The dataset also underlies the SCOLD vision-language model:
 
 ---
 
-This dataset is indexed and structured on https://project-agml.github.io/ as part of the AgML python library. This dataset was reformatted from its original format to match HuggingFace's Imagefolder standards but requires an external module (agml) that processes and returns a HF Dataset object faster than HF module functions.
+This dataset is indexed and structured on https://project-agml.github.io/ as part of the AgML python library. This dataset was reformatted from its original format to match HuggingFace's Imagefolder standards but requires an external module (agml) that processes and returns a HF Dataset object faster than HF module functions.