Real-time video generation demands fast decoding as much as fast denoising, yet current latent video diffusion models rely on 3D convolutional decoders that are slow and memory-intensive at high resolutions or for long video. We introduce FlashDecoder, a fast, memory-efficient pure-Transformer video decoder that decodes latents to pixels frame by frame. At each step, the current frame attends only to a fixed-size window of past frames through a rolling KV cache. The fixed temporal window keeps decoding fast and memory bounded regardless of video length, enabling constant-latency streaming. Because frames are processed sequentially, temporal causality is enforced without explicit attention masks, enabling training at resolutions up to 1080p and matching the reconstruction quality of convolutional decoders. On the Wan2.1 and Wan2.2 latent spaces, FlashDecoder matches each convolutional decoder in reconstruction quality (e.g., 41.55dB vs. 41.49dB PSNR at 1080p) while decoding 3.6x-4.7x faster with up to 11x less memory on a single H100 GPU. With architecture-aware inference optimizations, the speedup widens to 12x.
Reasoning temporal localization (RTL) requires a model to generate an answer that itself contains the time interval supporting it, coupling high-level reasoning with temporal grounding in a single response. To tackle this challenge, we propose the first event-centric video chain-of-thought framework, dubbed EventCoT. EventCoT first performs event-centric tokenization, converting the video into compact event tokens that enable efficient identification of question-relevant events. It then reasons within these events to generate the answer, grounding the time interval via embedding matching that aligns placeholder tokens with visual embeddings. EventCoT achieves state-of-the-art results on ActivityNet-RTL while using substantially fewer visual tokens than previous work, and attains strong zero-shot results on the grounded video question answering benchmark ReXTime. Our code will be released for research purposes.
Youngkil Song, Yoonjae Baek, Dongwon Kim et al.· 0 citations
This work revisits the efficacy of simple linear interpolation within an embedding space, and introduces SRAIN, the first framework that dynamically predicts query-specific interpolation weights, and achieves the best in composed video retrieval and matches the current state of the art in composed image retrieval.
Boseung Jeong, T. Park, Donghyeon Kwon et al.· 1 citation
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
