README.md

---
license: apache-2.0
tags:
- image-text-retrieval
- jepa
- contrastive-learning
- video-encoder
- pytorch
pipeline_tag: feature-extraction
---

# scout-eccv-v17

Part of **SCOUT** (Sim-to-real Contrastive Outlier Understanding via Text), an ECCV 2026 workshop
paper on JEPA-based text-based person re-identification for the AI City Challenge Track 4 (Sim2Real).
Code: https://github.com/abtraore/SCOUT-ECCV

Baseline SCOUT recipe: frozen V-JEPA2 ViT-L/16 (X) + frozen EmbeddingGemma-300M (Y) + a trainable 24-layer predictor, K=64 concat readout, per-forward batch 128.

- **Val mAP@10**: 0.8639
- **Val Hit@10**: 0.9862
- **Trainable / total parameters**: 549.6M / 1161.8M

The largest single ablation lever in the paper (per-forward InfoNCE batch, +0.094 mAP across the sweep) landed here. A fusion member of the paper's headline board system.

## What's in this repo

Only the parameters SCOUT actually trained: a `model.safetensors` (trainable predictor weights, and
for ExPLoRA/LoRA variants, the adapted encoder deltas) plus a `config.json` recording the exact recipe.
Frozen backbone weights (V-JEPA2 from Meta, the text encoder from its own source) are **not** included
here; they're downloaded from their own public sources at load time.

## Usage

```python
from scripts.load_from_hf import load_from_hf

model = load_from_hf("Abdrah/scout-eccv-v17").eval()
# model.encode_image(pixel_values)  # [B, 3, 384, 384] in [0, 1] -> [B, out_dim] L2-normalized
# model.encode_text(["a caption"])  # -> [B, out_dim] L2-normalized
```

Or drop it straight into the repo's own evaluation script:

```bash
uv run python scripts/local_eval.py --hf-repo Abdrah/scout-eccv-v17 --pab-root /path/to/PAB --query-source val
```

## License

Apache 2.0 (see the source repo's LICENSE/NOTICE). This checkpoint does not redistribute any frozen
backbone weights, which carry their own upstream licenses.