Paper Reviews by AI

SANA-Sprint: An efficient diffusion model for ultra-fast text-to-image generation with continuous-time consistency distillation, achieving state-of-the-art performance in speed and quality.

Reangle-A-Video generates synchronized multi-view videos from a single video via video-to-video translation, surpassing existing methods without specialized 4D training.

Quantization optimizes OpenAI’s Whisper models, balancing model size, speed, and accuracy for diverse applications.

PerCoV2: Open ultra-low bit-rate perceptual image compression using implicit hierarchical masked image modeling, built on Stable Diffusion 3 for bandwidth-constrained applications.

Open-Sora 2.0: A commercial-level video generation model trained for only $200k, achieving comparable results to state-of-the-art models.

VLMs struggle with basic image transforms! This paper reveals their limitations in understanding image-level changes, impacting downstream tasks.

NAR: Neighboring Autoregressive Modeling for efficient visual generation by locality-preserved, parallel decoding.

Group-robust machine unlearning via MIU reduces perf. degradation in dominant groups after unlearning, preserving model robustness without compromising accuracy.

Florenz: Scaling laws for systematic generalization via monolingual vision-language models

Unified framework reveals and mitigates error sources in autoregressive video diffusion models.

Cockatiel: Ensembling synthetic & human-preferred training boosts detailed video captioning, setting new SOTA on VDCSCORE.

UniFace: a novel UMM tailored for fine-grained face understanding and generation.

SynCoS: Synchronized sampling generates high-quality & coherent long videos from text, without extra training!

SegAgent: Improves MLLMs’ pixel understanding by mimicking human annotation, enabling mask refinement without altering output space.

Introducing HumanRef, a new dataset & RexSeek, a multimodal LLM, to improve human-centric referring tasks by addressing limitations of existing methods.

QuoTA: Task-aware token assignment boosts long video comprehension in LVLMs via query-decoupled processing, without extra training!

PLM retrievers overrate low-perplexity docs, causing source bias. This paper reveals the causal effect & offers a fix!

SBD: Self-supervised skill discovery from unsegmented videos!

OmniMamba: Efficient multimodal understanding and generation via SSMs, trained on 2M image-text pairs.

NullFace: A training-free face anonymization method preserving non-identity attributes with localized control using latent diffusion inversion.