IQAGA and DAPRH: UDA for Person ReID

• IQAGA and DAPRH are advanced unsupervised domain adaptation frameworks that use GAN-based augmentation to bridge the gap between source and target person re-ID data.
• IQAGA employs a two-stage StarGAN-based style transfer with image quality weighting to mitigate artifacts and enhance training stability.
• DAPRH integrates domain-invariant mapping, cluster-based pseudo-label refinement, and holistic ViT features, achieving significant mAP improvements on standard benchmarks.

IQAGA (Image Quality–Driven GAN Augmentation) and DAPRH (GAN Augmentation + Pseudo-Label Refinement + Holistic Features) are two advanced unsupervised domain adaptation (UDA) frameworks developed to address cross-domain generalization in person re-identification (ReID) tasks where source and target data distributions diverge sharply due to appearance variation, camera-specific styles, and lack of target labels. Both approaches utilize generative adversarial networks (GANs) for domain-specific image augmentation but diverge in loss composition, feature engineering, domain-invariant mapping, and pseudo-label mechanisms. Evaluated on standard ReID benchmarks, these frameworks show major improvements over prior UDA methods by systematically integrating augmentation, feature supervision, and robust target data exploitation (Pham et al., 4 Jan 2026).

1. IQAGA: Image Quality–Driven GAN Augmentation

IQAGA centers on a two-stage workflow combining StarGAN-based style transfer and image-quality-weighted supervised learning. In Stage I, StarGAN models are trained to convert each source image $x_s$ into $C_t$ styles corresponding to target camera domains, optimizing multiple objectives: adversarial ($L_{adv}$), domain classification ($L_{cls}$), cycle-reconstruction ($L_{rec}$), identity mapping ($L_{idt}$), and identity-preserving color loss ($L_{pid}$). Each source image's translations are concatenated with the original source set to construct a synchronized training set $D_{sync}$.

In Stage II, ResNet-50 provides 2048-D features per image. Supervised learning optimizes cross-entropy ($L_{ce}$) and triplet ($L_{tri}$) losses, but with per-sample IQA weighting leveraging normalized feature vector statistics ($C_t$0), directly modulating each image's gradient contribution: $C_t$1. Low-quality GAN samples thus exert reduced influence, mitigating mode collapse and spurious artifacts.

Key design choices in IQAGA include avoidance of target pseudo-labeling and domain-invariant mapping: adaptation is driven purely by GAN-based augmentation and image-level loss engineering.

2. DAPRH: GAN Augmentation, Pseudo-Label Refinement, Holistic Features

DAPRH expands the GAN augmentation paradigm by combining (i) a domain-invariant mapping (DIM) adversarial feature alignment, (ii) cluster-based pseudo-labeling with refinement, (iii) holistic feature encoding via Vision Transformer (ViT), and (iv) camera-aware proxies.

Stage I uses StarGAN for style transfer as in IQAGA, but batch construction incorporates a reduced ratio (e.g., 4:1 real:GAN), curbing GAN noise. DIM employs a domain classifier $C_t$2 to adversarially confound domain identity, training $C_t$3 to push $C_t$4 predictions toward 0.5 for both source and target features ($C_t$5).

In Stage II, target images are encoded, clustered by DBSCAN, and assigned hard pseudo-labels $C_t$6. Features are transformed by ViT/MLP, merging global (CLS token) and local MaxPool-selected top-K tokens for expressive queries $C_t$7. Cluster centers $C_t$8 support soft label refinement $C_t$9 (via softmax on Euclidean distance), further filtered by silhouette coefficient $L_{adv}$0. Refined labels take the weighted form $L_{adv}$1. A teacher-student framework employs EMA to update teacher weights, with student supervision via KL divergence ($L_{adv}$2) and soft triplet ($L_{adv}$3) losses.

DAPRH integrates camera-aware proxies by subdividing clusters via camera ID, yielding sub-centers $L_{adv}$4 and associated contrastive loss $L_{adv}$5 over proxies per sample. The unsupervised target-objective aggregates NCE, CAP, KL, and soft-triplet losses.

3. Mathematical Formulations

IQAGA Losses

• StarGAN Generator Loss:

$L_{adv}$6

• Discriminator Loss:

$L_{adv}$7

• IQA-weighted Source Loss:

$L_{adv}$8

DAPRH Losses (Stage I and II)

• DIM Loss:

$L_{adv}$9

• ClusterNCE:

$L_{cls}$0

• CAP Loss:

$L_{cls}$1

• Pseudo-label refinement:

$L_{cls}$2

• Teacher EMA update:

$L_{cls}$3

4. Training Protocols and Hyperparameterization

For both methods, StarGAN is trained on source versus target domain camera labels with Adam optimizer, learning rate $L_{cls}$4, batch size $L_{cls}$5, and loss weights $L_{cls}$6, $L_{cls}$7, $L_{cls}$8, $L_{cls}$9. ResNet-50 is initialized from ImageNet weights, supervised with Adam optimizer at $L_{rec}$0 (decayed at epochs $L_{rec}$1), batch size $L_{rec}$2, and $L_{rec}$3 epochs. Triplet margin $L_{rec}$4, IQA weight $L_{rec}$5.

DAPRH source batch formation reserves $L_{rec}$6 real:GAN images (e.g. $L_{rec}$7), uses $L_{rec}$8 for adversarial loss. Target clustering employs DBSCAN ($L_{rec}$9, MinPts=$L_{idt}$0 for Market, $L_{idt}$1 for MSMT), batch $L_{idt}$2. Top-K local tokens $L_{idt}$3 (approx 40% of total), $L_{idt}$4, $L_{idt}$5, $L_{idt}$6, $L_{idt}$7. Teacher EMA momentum $L_{idt}$8, $L_{idt}$9–$L_{pid}$0 epochs, SGD learning rate $L_{pid}$1.

5. Experimental Results and Ablation Findings

Quantitative results illustrate substantial gains in cross-domain scenarios:

Scenario	Baseline mAP / Rank-1	+GAN	+GAN+IQA / +DIM	DAPRH Final
Market→Duke	25.8 / 43.7	31.5 / 54.2	32.1 / 55.5	72.0 / 83.7
Duke→Market	26.2 / 55.3	35.1 / 68.6	36.3 / 70.2	85.9 / 94.4
Market→MSMT	—	—	—	35.8 / 64.8
Duke→MSMT	—	—	—	36.0 / 65.5

Ablation analyses show:

• GAN augmentation alone yields $L_{pid}$2 mAP improvement; IQA weighting adds another $L_{pid}$3 mAP in IQAGA.
• In DAPRH, DIM is more computationally efficient than GAN for early-stage feature alignment, but integrating both is optimal.
• Holistic (ViT) features and CAP each contribute $L_{pid}$4–$L_{pid}$5 mAP; jointly, a further $L_{pid}$6–$L_{pid}$7 mAP.
• CRL and teacher-student boost $L_{pid}$8–$L_{pid}$9 mAP, crucial for scaling to large datasets.
• Key hyperparameters exhibit clear optima: $D_{sync}$0–$D_{sync}$1, $D_{sync}$2, top-K $D_{sync}$3.

This suggests that high-fidelity augmentation, loss weighting, domain confusion, and advanced pseudo-labeling together address critical bottlenecks in fully unsupervised ReID adaptation.

6. Contributions and Comparative Significance

IQAGA demonstrates that simple GAN-based augmentation, when augmented with IQA-driven sample weighting, surpasses prior GAN-based UDA by $D_{sync}$4–$D_{sync}$5 mAP. DAPRH incorporates multi-component alignment—style transfer, adversarial mapping, refined soft pseudo-labels, holistic feature representation, camera-aware local proxies—and achieves more than $D_{sync}$6 mAP on Market→Duke and $D_{sync}$7 mAP on Duke→Market, bridging much of the practical gap to fully supervised approaches. On large-scale MSMT, DAPRH exceeds $D_{sync}$8 mAP, surpassing previous unsupervised adaptation results.

A plausible implication is that multi-stage integration of style transfer, discriminative feature enhancement, cluster-based label refinement, and domain-invariant mapping forms an effective paradigm for cross-domain ReID without target labels. Further, the critical role of image quality assessment and proxy learning highlights the importance of robust sample and feature selection in deep UDA pipelines, a point of emerging significance for unsupervised visual recognition research (Pham et al., 4 Jan 2026).