DeepSketcher supplies a 31k interleaved reasoning dataset and a VLM that internalizes visual manipulations by operating natively in visual embedding space to enable tool-free thinking with images.
thinking with images
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
DeepSketcher: Internalizing Visual Manipulation for Multimodal Reasoning
DeepSketcher supplies a 31k interleaved reasoning dataset and a VLM that internalizes visual manipulations by operating natively in visual embedding space to enable tool-free thinking with images.