Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Perceptio: Perception Enhanced Vision Language Models via Spatial Token Generation

About

Large Vision Language Models (LVLMs) excel at semantic understanding but struggle with fine grained spatial grounding, as the model must implicitly infer complex geometry without ever producing a spatial interpretation. We present Perceptio, a perception enhanced LVLM with 2D and 3D spatial reasoning abilities, enabled via explicit semantic segmentation tokens and depth tokens generated directly within the autoregressive sequence. Concretely, we (i) distill a VQVAE depth codebook from a strong monocular teacher to tokenize dense depth into compact sequences, and (ii) integrate SAM2 based semantic segmentation tokens and VQ-VAE depth tokens inside the LLM so the model first emits spatial tokens and then answers. To stabilize depth token generation, we introduce novel composite depth-token objectives (marker, token, and count losses) and a soft-merging technique for differentiable reconstruction. We adopt a multi-task co-training strategy across diverse datasets, letting the model learn perception tokens to tackle multiple downstream tasks. Building on InternVL, Perceptio achieves state-of-the-art performance across benchmarks: improving referring expression segmentation by +0.8/+1.4/+1.1 cIoU on RefCOCO/+/g HardBLINK spatial understanding accuracy by 10.3%, and MMBench accuracy by 1.0%, demonstrating that explicit spatial chain-of-thought materially strengthens spatial grounding in LVLMs.

Yuchen Li, Amanmeet Garg, Shalini Chaudhuri, Rui Zhao, Garin Kessler• 2026

Related benchmarks

TaskDatasetResultRank
Diagram UnderstandingAI2D
Accuracy83.4
247
Science Question AnsweringScienceQA (test)
Average Accuracy98.3
245
Image-Text UnderstandingMMBench
Accuracy83.4
32
Referring Image SegmentationRefCOCO
cIoU82.7
11
Referring Image SegmentationRefCOCO+
cIoU77.9
11
Referring Image SegmentationRefCOCOg
cIoU80
11
Image Chat / UnderstandingMME
MME-T Score2.33e+3
9
Relative Depth ReasoningHardBLINK
Accuracy (3 Points)75.8
7
Image Chat / UnderstandingSEED-Bench
Accuracy75.7
7
Image Chat / UnderstandingMMStar
Accuracy64.2
4
Showing 10 of 10 rows

Other info

Follow for update