## How is testing done
The AG-VPReID 2025 competition evaluates submissions by computing mean Average Precision (mAP) (auto) and Cumulative Matching Characteristics (CMC) (manual) at Rank-1, Rank-5, and Rank-10. Participants' algorithms are tested on video tracklets from high-altitude aerial footage (80m-120m) and ground cameras, featuring individuals not included in the training sample. For each query tracklet, the gallery tracklets must be ranked based on similarity, with the query itself excluded from matching results.

The evaluation focuses specifically on two challenging cases:
1. Using aerial tracklets as queries to find matching ground tracklets
2. Using ground tracklets as queries to find matching aerial tracklets


Baseline code and implementation guides will be made available through our GitHub repository [link](https://github.com/agvpreid25/AG-VPReID)

## Metrics
This competition employs the following metrics:
- Mean Average Precision (mAP): Evaluates the overall ranking quality across all queries
- Cumulative Matching Characteristics (CMC) at Rank-1, Rank-5, and Rank-10: Measures the probability of finding the correct match within the top-k results

## Submission File
Participants must provide a submission file containing the predicted gallery tracklets for each query tracklet. Each submission should follow this format:

```
row_id,query_tracklet,gallery_tracklets
1,query_tracklet_id_1,gallery_tracklet_id_1 gallery_tracklet_id_2 ... gallery_tracklet_id_10
2,query_tracklet_id_2,gallery_tracklet_id_3 gallery_tracklet_id_4 ... gallery_tracklet_id_11
```

Where:
- `row_id`: A unique identifier for each row
- `query_tracklet`: The ID of the query tracklet
- `gallery_tracklets`: A space-separated list of up to 10 gallery tracklet IDs, ranked by similarity (highest first)

## Note
>**IMPORTANT**: The person identity labels in the gallery set are hidden from participants, meaning participants will not know whether their algorithm has correctly identified the target person until they submit their results and receive evaluation feedback. Participants may submit up to 10 times / day, with the best performance considered as their final result. External training data may be used but must be disclosed, and such submissions will be ranked separately.