Saved in:
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2605.17260 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866910252124864512 |
|---|---|
| author | Kim, Jihwan Parthasarathy, Nikhil Qin, Danfeng Hur, Junhwa Sun, Deqing Han, Bohyung Yang, Ming-Hsuan Gong, Boqing |
| author_facet | Kim, Jihwan Parthasarathy, Nikhil Qin, Danfeng Hur, Junhwa Sun, Deqing Han, Bohyung Yang, Ming-Hsuan Gong, Boqing |
| contents | The fundamental challenge in scaling Video Large Language Models (Video LLMs) to long-form video lies in managing the explosion of visual-token context length. Existing strategies predominantly focus on "post-hoc" token reduction -- reducing visual tokens after feature extraction to alleviate the LLM's computational overhead. While these methods effectively reduce the number of visual tokens, we observe that the primary latency bottleneck then shifts from the LLM to the expensive per-frame processing of the vision encoder. To address this, we introduce LiteFrame, a strong, yet highly efficient video encoder backbone for Video LLMs. To train LiteFrame, we propose Compressed Token Distillation (CTD), a novel training framework that teaches a compact student vision encoder to directly predict information-dense, spatio-temporally compressed representations produced by a large teacher vision model, effectively bypassing redundant computation. When coupled with further Language Model Adaptation (LMA), this approach results in a new latency-accuracy Pareto frontier -- compared with InternVL3-8B, LiteFrame provides a 35% reduction in end-to-end latency while processing 8$\times$ more frames and improves average video understanding accuracy across multiple benchmarks. Our results demonstrate a new potential path to unlocking longer-form video understanding under fixed compute budgets. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_17260 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | LiteFrame: Efficient Vision Encoders Unlock Frame Scaling in Video LLMs Kim, Jihwan Parthasarathy, Nikhil Qin, Danfeng Hur, Junhwa Sun, Deqing Han, Bohyung Yang, Ming-Hsuan Gong, Boqing Computer Vision and Pattern Recognition The fundamental challenge in scaling Video Large Language Models (Video LLMs) to long-form video lies in managing the explosion of visual-token context length. Existing strategies predominantly focus on "post-hoc" token reduction -- reducing visual tokens after feature extraction to alleviate the LLM's computational overhead. While these methods effectively reduce the number of visual tokens, we observe that the primary latency bottleneck then shifts from the LLM to the expensive per-frame processing of the vision encoder. To address this, we introduce LiteFrame, a strong, yet highly efficient video encoder backbone for Video LLMs. To train LiteFrame, we propose Compressed Token Distillation (CTD), a novel training framework that teaches a compact student vision encoder to directly predict information-dense, spatio-temporally compressed representations produced by a large teacher vision model, effectively bypassing redundant computation. When coupled with further Language Model Adaptation (LMA), this approach results in a new latency-accuracy Pareto frontier -- compared with InternVL3-8B, LiteFrame provides a 35% reduction in end-to-end latency while processing 8$\times$ more frames and improves average video understanding accuracy across multiple benchmarks. Our results demonstrate a new potential path to unlocking longer-form video understanding under fixed compute budgets. |
| title | LiteFrame: Efficient Vision Encoders Unlock Frame Scaling in Video LLMs |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2605.17260 |