Visual Inference Lab

In-context learning suggests that a model has learned versatile representations. What if we use in-context learning itself as a training task for visual representations? 📣 Introducing 𝗟𝗜𝗟𝗔: 𝗟𝗶𝗻𝗲𝗮𝗿 𝗜𝗻-𝗖𝗼𝗻𝘁𝗲𝘅𝘁 𝗟𝗲𝗮𝗿𝗻𝗶𝗻𝗴 ✨ @CVPR 2026 Oral ✨ 𝗟𝗜𝗟𝗔 trains on videos without manual annotation. Key idea: An optimal linear mapping that predicts dense cues (e.g. depth, flow), estimated on one video frame, should also predict the corresponding cues of another frame from the same video. This yields compelling results on dense vision tasks: video object segmentation, (zero-shot) semantic segmentation and surface normal estimation. Paper, code, models and demo: lila-pixels.github.io Joint work with @ma_sundermeyer, Hidenobu Matsuki, David Joseph Tan and @fedassa (and special thanks to David and Federico for hosting my research visit at Google). #cvpr2026 @Google @MunichCenterML @tumcvg @TU_Muenchen

📢 NeurIPS 2025 Spotlight 📢 Can we embed motion into image representations? Trained on videos, FlowFeat embeds optical flow into pixel-level representations (up to a linear transform), which results in sharp feature grids, especially for dynamic objects. We demonstrate benefits for ⭐️video object segmentation; ⭐️semantic segmentation; ⭐️and monocular depth. Paper: arxiv.org/abs/2511.07696 Project website: tum-vision.github.io/flowfea… Code and models: github.com/tum-vision/flowfe… Joint work with Anna Sonnweber and Daniel Cremers @tumcvg and @MunichCenterML. Come by our poster @NeurIPSConf on Thursday (Exhibit Hall C,D,E #4816)!