Domain Adaptation & Generalization
Cold-Start Streaming Domain Adaptation and Discovery Without Labeled Base Sessions or Offline Pre-Passes
Generated automatically from the limitations stated in 3 papers (CVPR, ICLR), listed under Evidence. It is not a paper, and it does not come from papers submitted to CSPaper.
The problem
Current continual domain adaptation and streaming open-world discovery methods structurally depend on an offline initialization phase—either requiring a fully labeled source base dataset or an offline full-dataset inference pass to compute initial prototypes and representations. In truly dynamic environments where data arrives purely as an online stream and domain shifts occur from the first observation, collecting labeled base data or executing multi-pass offline initialization is impossible. Consequently, existing frameworks cannot be deployed in pure cold-start streaming regimes.
Why it matters
Enables autonomous deployment of streaming adaptation and class discovery models from time zero without prior source data collection, manual labeling, or batch target indexing.
Ways to approach it
Prior-work checks are free with an account. Results someone already ran are shown to everyone.
- 1
Online prototype bootstrapping via self-supervised clustering on initial streaming buffers, measuring early-stream error rates and convergence speed of prototype stability across standard domain adaptation benchmarks.
- 2
Dynamic density-based prototype initialization using zero-shot foundation model features without offline target passes, measuring discovery accuracy and catastrophic forgetting under continuous domain shift.
- 3
Multi-fidelity online pseudo-labeling that adaptively tunes cluster assignment thresholds as streaming sample volume grows, evaluated against offline-initialized upper bounds on streaming domain benchmarks.
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 pure online initialization suffers from irreversible early-stage confirmation bias and error propagation that cannot be corrected without at least a minimal curated offline warmup set.
Evidence
Each paper's own statement of the limitation, verbatim.
- PRISM: Progressive Robust Learning for Open-World Continual Category DiscoveryICLR 2026
Requires a fully labeled, single-domain base session to establish initial prototypes and classifier representations prior to continual streaming.
- Expert-Teacher-Student Collaborative Learning for Domain Adaptive Object DetectionCVPR 2026
Requires generating offline pseudo-labels and prototypes via an initial full forward pass over the target dataset using large foundation models before mutual training.
- Beyond the Static World: Continual Category Discovery under Visual DriftCVPR 2026
Requires a fully labeled base-session dataset from a known source domain for initial prototype and representation learning prior to streaming discovery.
Nearest existing work
- Prototype-Guided Continual Adaptation for Class-Incremental Unsupervised Domain AdaptationECCV 2022
- Generalized Incremental Learning under Concept Drift across Evolving Data StreamsTheWebConference 2026
- Continual Prototype Evolution: Learning Online From Non-Stationary Data StreamsICCV 2021
- Exemplar-free Continual Representation Learning via Learnable Drift CompensationECCV 2024
- Unsupervised Continual Domain Shift Learning with Multi-Prototype ModelingCVPR 2025
- TALON: Test-time Adaptive Learning for On-the-Fly Category DiscoveryCVPR 2026
- Online Prototype Learning for Online Continual LearningICCV 2023
- Online Domain Adaptation for Semantic Segmentation in Ever-Changing ConditionsECCV 2022
- Learning from Snapshots of Discrete and Continuous Data StreamsNeurIPS 2024
- Burn after Reading: Online Adaptation for Cross-Domain Streaming DataECCV 2022
- Adaptive Methods for Real-World Domain GeneralizationCVPR 2021
- Beyond the Static World: Continual Category Discovery under Visual DriftCVPR 2026
- Resurrecting Old Classes with New Data for Exemplar-Free Continual LearningCVPR 2024
- Category Adaptation Meets Projected Distillation in Generalized Continual Category DiscoveryECCV 2024
- Reshaping the Online Data Buffering and Organizing Mechanism for Continual Test-Time AdaptationECCV 2024
Related open problems
Domain Adaptation & Generalization
Domain-Invariant Learning Without Domain Labels or Target-Domain Data at Training Time
Current methods across these settings all condition on information that disappears in deployment: explicit domain labels for training samples, knowledge of the domain count, or a pool of unlabeled target-domain data available during training. When source domain identity is unknown, only one source is available, or target data cannot be collected before the model is fixed, these methods are simply unusable — not merely degraded. This blocks the realistic scenario where a single labeled dataset is all one has, and the test distribution is only discovered at inference time.
Domain Adaptation & Generalization
Cross-Architecture Robustness and Evaluation of Domain Adaptation in Object Detection
Current domain adaptation and domain generalization methods for object detection are almost exclusively evaluated on two-stage anchor-based architectures, primarily Faster R-CNN with ResNet backbones. Because these adaptation mechanisms often couple directly to region proposal networks, RoI pooling representations, or batch normalization layers, it is completely unknown whether reported domain adaptation gains transfer to one-stage detectors (e.g., FCOS, YOLO) or query-based transformers (e.g., DETR variants). Consequently, practitioners deploying modern transformer or anchor-free architectures have no empirical guidance on whether established domain adaptation methods function outside Faster R-CNN.
Domain Adaptation & Generalization
Systematic Evaluation of Data Augmentation Techniques Across Disparate Distribution Shift Types
Data augmentation methods developed for out-of-distribution robustness are overwhelmingly hand-engineered to simulate surface-level texture and appearance variations (e.g., color perturbations and synthetic corruptions). When these methods are deployed under other distribution shift types—such as subpopulation shifts, geometric transformations, or domain shifts in specialized fields like medical imaging—they frequently fail or degrade accuracy below standard Empirical Risk Minimization (ERM). Because augmentation techniques have not been systematically evaluated across these differing shift categories, practitioners have no reliable way to anticipate when a given augmentation strategy will harm downstream generalization.