Person Re-Identification & Face Recognition
Cross-Distribution Robustness and Granularity Evaluation in Cloth-Changing Person Re-Identification
Generated automatically from the limitations stated in 5 papers (CVPR, NeurIPS), listed under Evidence. It is not a paper, and it does not come from papers submitted to CSPaper.
The problem
Existing cloth-changing person re-identification (CC-ReID) and gait recognition models are trained and tested on datasets with restricted clothing variation profiles, such as upper-body-only changes or invariant footwear. Models trained on these biased distributions experience severe performance drops (such as single-digit Rank-1 accuracy) when evaluated on full clothing or pants-only changes. Without a standardized cross-setting evaluation that separates upper, lower, full, and footwear changes, reported benchmark metrics obscure localized overfitting and fail to measure true clothing-invariant representations.
Why it matters
Establishes a rigorous diagnostic framework to identify which feature representations achieve true clothing invariance rather than exploiting static regional artifacts like unchanged footwear.
Ways to approach it
Prior-work checks are free with an account. Results someone already ran are shown to everyone.
- 1
Multi-partition benchmark evaluation: Re-categorize test splits across existing benchmarks (e.g., CCPG, LTCC, PRCC) into distinct clothing-change partitions (upper-only, pants-only, full-change, shoe-masked vs. unmasked) to measure Rank-1 and mAP across leading RGB, gait, and disentanglement methods.
- 2
Cross-dataset distribution shift study: Train representative CC-ReID models on datasets with varying clothing change distributions (e.g., SUSTech1K, CASIA-B*, and CCGR_MINI) and evaluate zero-shot cross-dataset generalization across varying clothing change granularities.
- 3
Supervision sensitivity profiling: Measure the performance degradation of label-free/unsupervised, text-guided, and explicitly annotated multi-branch methods across varying ratios of clothing-change training pairs.
Have a different approach?
Describe how you would tackle this problem and we'll look for papers that already do it. Free; your text stays private.
Why it might fail
If trivial random masking augmentations during training completely resolve the cross-partition degradation across all existing architectures, rendering a dedicated robustness benchmark redundant.
Evidence
Each paper's own statement of the limitation, verbatim.
- Composite-Attribute Person Re-Identification via Pose-Guided DisentanglementCVPR 2026
On standard CC-ReID benchmarks the method underperforms specialized clothes-changing methods (63.8 vs 68.5 Top-1 on LTCC vs DIFFER), so it does not replace them
- BiggerGait: Unlocking Gait Recognition with Layer-wise Representations from Large Vision ModelsNeurIPS 2025
Training-data distribution bias: trained on CCGR_MINI and tested on CCPG, results degrade on full-clothing (CL) and pants-changing (DN) conditions, attributed to limited clothing diversity (only 9.4% pairs, exclusively UP changes)
- BigGait: Learning Gait Representation You Want by Large Vision ModelsCVPR 2024
Performance is heavily dependent on training data distribution: trained on SUSTech1K (<5% cross-dressing pairs), BigGait fails on cross-dressing settings (4.5 CL rank-1 on CCPG); trained on CASIA-B* (ups-only changes) it does poorly on full-changing (CL) and pants-changing (DN)
- Good Is Bad: Causality Inspired Cloth-Debiasing for Cloth-Changing Person Re-IdentificationCVPR 2023
Requires clothing/suit annotations and sufficient samples to train the clothing branch, limiting applicability where such labels are absent
- An In-Depth Exploration of Person Re-Identification and Gait Recognition in Cloth-Changing ConditionsCVPR 2023
Shoes were unchanged across outfits due to collection constraints, so ReID performance on the unmasked subsets (CCPG-A/B) is inflated; only the shoe-masked subsets approximate real cloth changing
Nearest existing work
- Clothes-Changing Person Re-Identification With RGB Modality OnlyCVPR 2022
- An In-Depth Exploration of Person Re-Identification and Gait Recognition in Cloth-Changing ConditionsCVPR 2023
- Clothing Status Awareness for Long-Term Person Re-IdentificationICCV 2021
- Cloth-Changing Person Re-Identification From a Single Image With Gait Prediction and RegularizationCVPR 2022
- COCAS: A Large-Scale Clothes Changing Person Dataset for Re-IdentificationCVPR 2020
- DIFFER: Disentangling Identity Features via Semantic Cues for Clothes-Changing Person Re-IDCVPR 2025
- BarbieGait: An Identity-Consistent Synthetic Human Dataset with Versatile Cloth-Changing for Gait RecognitionCVPR 2026
- Colors See Colors Ignore: Clothes Changing ReID with Color DisentanglementICCV 2025
- Identity-Clothing Similarity Modeling for Unsupervised Clothing Change Person Re-IdentificationCVPR 2025
- Learning Clothing and Pose Invariant 3D Shape Representation for Long-Term Person Re-IdentificationICCV 2023
- Good Is Bad: Causality Inspired Cloth-Debiasing for Cloth-Changing Person Re-IdentificationCVPR 2023
- Towards Realistic Lifelong Re-identification: Identity Recurrence with Changing ClothesICML 2026
- DeepChange: A Long-Term Person Re-Identification Benchmark with Clothes ChangeICCV 2023
- Fine-Grained Shape-Appearance Mutual Learning for Cloth-Changing Person Re-IdentificationCVPR 2021
- HAMoBE: Hierarchical and Adaptive Mixture of Biometric Experts for Video-based Person ReIDICCV 2025
Related open problems
Person Re-Identification & Face Recognition
Robust Cross-Modality and Cross-Platform Retrieval for Person and Vehicle Re-Identification
Current cross-modality re-identification methods report relative improvements on benchmark datasets, yet their absolute retrieval performance drops severely in strongly mismatched sensor and platform scenarios (e.g., 9–16 mAP on RGBNT201 and MSVR310). When spectral shifts (RGB to Near-Infrared or Thermal) coincide with cross-platform viewpoint or domain shifts, identity discrimination fails at a fundamental level. This severe degradation blocks the practical deployment of automated re-identification systems across heterogeneous, multi-sensor surveillance networks operating in varying lighting and platform conditions.
Person Re-Identification & Face Recognition
Cross-Encoder and Target-Free Robustness Benchmarking for Facial Representations
Current facial representation and transfer methods rely on training-time preconditions that frequently fail in real-world deployment. Specifically, techniques depend on collecting unlabeled target-domain demographic data—which is often legally or ethically prohibited—or evaluate strictly against known, seen feature extractors within face-only datasets. When the deployed encoder is changed or target data cannot be sampled in advance, systems experience substantial performance drops that remain unquantified across modern architectures. Without a unified benchmark testing methods under zero-target-data and unseen-encoder constraints, practitioners cannot determine which representation strategies legitimately transfer.