CritPt
Harbor: pendingCritPt — AI evaluation benchmark listed by opencompass.
Verified71%Knowledge & Reasoning
- Samples
- 70
- Released
- 2025
- Scores
- 138
Front2Back-ReID
Front-to-Back: Benchmarking Vision-Language Models for Asymmetric Cross-View Vehicle Re-Identification
A benchmark of 500 manually verified vehicle handovers from South African roads evaluating zero-shot VLMs, frozen retrieval baselines, and humans on front-to-rear matching using full RGB scenes, crops, or silhouettes.
How to run
ltb · front2back-reid
$ npm i -g lettokenburn-cli$ ltb bench info front2back-reid
External harness. No Harbor adapter has been generated for this benchmark yet. `ltb bench info` prints the known links.
Runs happen on your machine with your own subscription; only the score summary is pushed. CLI docs
Scores
3 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| GPT-5.5 | 62.8 | N/A | Independent | medium | Sep 30, 2026 | arxiv.org |
| Gemini 2.5 Pro | 60.6 | N/A | Independent | medium | Sep 30, 2026 | arxiv.org |
| GPT-5.4-mini | 58.8 | N/A | Independent | medium | Sep 30, 2026 | arxiv.org |
Every number links to where it was published. Low-confidence rows are muted and need per-cell verification.
Reproduced by the community
All runsLoading community runs…
About
Content & Construction: The dataset consists of 500 asymmetric cross-view vehicle re-identification (Re-ID) handovers recorded from 20 sequences in South Africa. It features low-cost stereo cameras mounted on an ego vehicle to capture front and rear views with a blind interval between observations, creating challenging distractors.
Scoring & Key Findings: Models are evaluated by selecting the unique matching candidate among at least three options (closed gallery). The strongest Vision-Language Model achieved 76.6% Rank-1 accuracy on target crops, while humans reached 94.0%. Results indicate that general-purpose VLMs do not yet consistently outperform strong visual retrieval baselines like SigLIP2.