MIPR / data /README.md
leequeue's picture
Upload folder using huggingface_hub
49654fc verified
|
Raw History Blame Contribute Delete
1.36 kB
# MIPR data
| File / directory | Contents |
|---|---|
| `mipr_query.jsonl` | Queries: `idx`, original `query`, and local image filename `image_url` |
| `mipr_corpus.jsonl` | Products: `item_id`, `title`, local image filename `pict_url`, and leaf category path `treecate_full_path` |
| `mipr_annotation.jsonl` | Pooled five-grade relevance judgments, grouped by query |
| `images/query_images/` | Query images referenced by `image_url` |
| `images/product_images/` | Product images referenced by `pict_url` |
| `baseline/` | Saved top-50 rankings for retrieval and reranking evaluation |
## Baseline rankings
`baseline/qwen3_vl_embedding_8b_top50.json` contains the top-50 retrieval results from Qwen3-VL-Embedding-8B. `baseline/qwen3_vl_reranker_8b_top50.json` contains the same candidate pool reordered by Qwen3-VL-Reranker-8B.
These are JSON rankings, not model weights or embedding arrays. Each file contains 3,522 query entries with `qid` and `ranked`; each ranked item has `did` and `score`. They allow evaluation of the saved results without running model inference.
From the release root:
```bash
python evaluation/evaluate.py data/baseline/qwen3_vl_embedding_8b_top50.json
python evaluation/evaluate.py data/baseline/qwen3_vl_reranker_8b_top50.json --rerank
```
See the [main README](../README.md) for data schemas, relevance grades and metrics.