Update README.md

README.md

@@ -11,14 +11,19 @@ DISBench is the first benchmark for context-aware image retrieval over visual hi
 
 ## Download
 
+**Option A: Hugging Face (Recommended)**
+
 ```bash
-# Option 1: Hugging Face
-huggingface-cli download xxx/DISBench --local-dir .
+huggingface-cli download RUC-NLPIR/DISBench --local-dir DISBench
+```
 
-# Option 2: Download images from YFCC100M
+**Option B: Manual Download**
+
+```bash
 python download_images.py --photo-ids-path photo_ids --images-path images
 ```
 
+
 ## File Structure
 
 ```
@@ -31,7 +36,6 @@ DISBench/
 │       └── {photo_id}.jpg     # Photo files
 ├── photo_ids/
 │   └── {user_id}.txt          # Photo IDs and hashes per user
-├── evaluate.py                # Evaluation script
 └── download_images.py         # Image download script
 ```
 
@@ -43,10 +47,10 @@ Each line is a JSON object representing one query:
 
 ```json
 {
-    "query_id": "q001",
-    "user_id": "12345678",
+    "query_id": "1",
+    "user_id": "10287726@N02",
     "query": "Find photos from the musical performance identified by the blue and white event logo on site, where only the lead singer appears on stage.",
-    "answer": ["98765432", "98765433", "98765434"],
+    "answer": ["7759256930", "7759407170", "7759295108", "7759433016"],
     "event_type": "intra-event"
 }
 ```
@@ -65,14 +69,14 @@ Each line is a JSON object representing one photo's metadata:
 
 ```json
 {
-    "photo_id": "98765432",
+    "photo_id": "4517621778",
     "metadata": {
-        "taken_time": "2012-08-03 14:32:10",
-        "longitude": -1.8808,
-        "latitude": 50.7192,
+        "taken_time": "2010-04-10 13:52:57",
+        "longitude": -1.239802,
+        "latitude": 51.754123,
         "accuracy": 16.0,
-        "address": "Bournemouth, Dorset, England",
-        "capturedevice": "Canon EOS 550D"
+        "address": "Y, Cherwell Street, St Clement's, East Oxford, Oxford, Oxfordshire, England, OX4 1BQ, United Kingdom",
+        "capturedevice": "Panasonic DMC-TZ5"
     }
 }
 ```
@@ -105,119 +109,6 @@ Each line represents one photo ID and its hash on aws storage in the format `{ph
 | `photo_id` | string | Unique photo identifier |
 | `hash` | string | Hashed value of the photo on aws storage |
 
-## Evaluation
-
-### Agent Evaluation
-
-For agent systems that predict a set of photo IDs per query:
-
-```bash
-python evaluate.py \
-    --mode agent \
-    --dataset_path . \
-    --prediction_path /path/to/predictions.jsonl \
-    --output_dir /path/to/run_id_dir/
-```
-
-**Prediction format** — a JSONL file where each line is:
-
-```json
-{
-    "query_id": "q001",
-    "prediction": ["98765432", "98765433"]
-}
-```
-
-**Output files:**
-
-`eval_samples.jsonl` — per-query results:
-
-```json
-{
-    "query_id": "q001",
-    "user_id": "12345678",
-    "query": "Find photos from the musical performance...",
-    "answer": ["98765432", "98765433", "98765434"],
-    "prediction": ["98765432", "98765433"],
-    "tp": 2, "fp": 0, "fn": 1,
-    "em": 0,
-    "iou": 0.667,
-    "precision": 1.0,
-    "recall": 0.667,
-    "f1": 0.8
-}
-```
-
-`eval_summary.json` — aggregated results:
-
-```json
-{
-    "n_samples": {
-        "total": 122,
-        "inter_event": 65,
-        "intra_event": 57
-    },
-    "aggregation": {
-        "total": {
-            "macro": { "em": 0.271, "iou": 0.45, "precision": 0.55, "recall": 0.52, "f1": 0.513 },
-            "micro": { "tp": 230, "fp": 45, "fn": 60, "iou": 0.42, "precision": 0.836, "recall": 0.793, "f1": 0.814 }
-        },
-        "inter_event": { "macro": { "..." : "..." }, "micro": { "..." : "..." } },
-        "intra_event": { "macro": { "..." : "..." }, "micro": { "..." : "..." } }
-    },
-    "set_size": {
-        "avg_answer_size": { "total": 3.76, "inter_event": 3.52, "intra_event": 4.04 },
-        "avg_pred_size": { "total": 3.10, "inter_event": 2.80, "intra_event": 3.44 }
-    },
-    "notes": {
-        "empty_pred_count": { "total": 5, "inter_event": 3, "intra_event": 2 }
-    }
-}
-```
-
-### Retriever Baseline Evaluation
-
-For embedding-based retrieval that returns a ranked list:
-
-```bash
-python evaluate.py \
-    --mode retriever \
-    --dataset_path . \
-    --prediction_path /path/to/predictions.jsonl \
-    --output_dir /path/to/retriever_dir/
-```
-
-**Prediction format** — a JSONL file where each line is:
-
-```json
-{
-    "query_id": "q001",
-    "prediction": ["photo_rank1", "photo_rank2", "photo_rank3", "..."]
-}
-```
-
-> Predictions should be ordered by descending relevance score.
-
-**Output files:**
-
-`eval_samples.jsonl` — per-query results with MAP@k, Recall@k, NDCG@k for k ∈ {1, 3, 5, 10}.
-
-`eval_summary.json` — aggregated metrics:
-
-```json
-{
-    "n_samples": 122,
-    "aggregation": {
-        "MAP@1": 0.123, "MAP@3": 0.105, "MAP@5": 0.120, "MAP@10": 0.133,
-        "Recall@1": 0.054, "Recall@3": 0.116, "Recall@5": 0.170, "Recall@10": 0.250,
-        "NDCG@1": 0.123, "NDCG@3": 0.133, "NDCG@5": 0.157, "NDCG@10": 0.188
-    },
-    "notes": {
-        "incomplete_pred_count": 0
-    }
-}
-```
-
 ## Dataset Statistics
 
 | Statistic | Value |
@@ -227,7 +118,7 @@ python evaluate.py \
 | Inter-Event Queries | 65 (53.3%) |
 | Total Users | 57 |
 | Total Photos | 109,467 |
-| Avg. Targets per Query | 3.76 |
+| Avg. Targets per Query | 3.84 |
 | Avg. History Span | 3.4 years |
 | Query Retention Rate | 6.1% (122 / 2,000 candidates) |
 | Inter-Annotator IoU | 0.91 |