Precomputed evaluation and benchmark results for Visual Product Search
Indexed Images
Evaluation Queries
Embedding Dimension
Self-Match Excluded
Article Type Precision@5
Subcategory Precision@5
Gender + Master Category Precision@5
Model: mohsin416/fashion-clip-vit-large | Dataset: paramaggarwal/fashion-product-images-dataset
| Match Type | Definition | Precision@5 | NDCG@10 | mAP@10 | MRR@10 |
|---|---|---|---|---|---|
| Article Type Match | Same articleType | 86.32% | 85.07% | 79.75% | 91.57% |
| Subcategory Match | Same subCategory | 96.4% | 95.95% | 94.48% | 97.87% |
| Gender + Master Category Match | Same gender and masterCategory | 90.96% | 90.39% | 87.04% | 94.68% |
Search latency measured using precomputed embeddings.
Mean Latency
P50 Latency
P95 Latency
P99 Latency
| Top-K | 10 |
| Measured Runs | 100 |
| Embedding File Size | 129.0968 MB |
| Metadata File Size | 16.4993 MB |