Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces Continuous Thought Machines (CTM) to enhance image super-resolution by incorporating an internal temporal dimension that evolves neuron-level histories and representations. The method, termed ThinkSR, utilizes a window-level approach where compact thought representations are transformed into dense queries, allowing for localized refinement of visual data without pooling. Preliminary results show significant improvements in image quality metrics, with PSNR values increasing notably across training ticks, demonstrating the potential of sparse latent thought for dense visual tasks.
Sparse latent thought can dramatically enhance image super-resolution, achieving substantial quality improvements in high-resolution reconstructions.
Continuous Thought Machines introduce an internal temporal dimension in which neuron-level histories and synchronization-derived representations evolve over a sequence of thought ticks. Extending this mechanism to dense visual prediction is non-trivial, because tasks such as image super-resolution require spatial evidence to remain available at every output location rather than being compressed into a single global representation. In the proposed window-level use of CTM, the thought dynamics produce a compact summary representation for each local window. DQ-CTM transforms this compact thought representation into window-aligned dense queries through a structured low-rank, parameter-efficient compact-to-dense query mechanism. Each position within a window receives its own query, while shared thought dynamics progressively refine the dense representation across ticks. In its super-resolution instantiation, termed ThinkSR, encoded feature maps are partitioned into local visual windows without token pooling, restored to the original feature field after shared refinement, and decoded into a high-resolution image. Preliminary experiments under a fixed four-tick training horizon reveal a progressive reconstruction trajectory. PSNR-Y increases from 28.1045 dB at $T=0$ to 30.2817 dB at $T=4$, while PSNR-RGB increases from 26.6271 dB to 28.7781 dB and the mean $\ell_1$ error decreases from 0.034602 to 0.023545. All 100 evaluated images improve from $T=1$ to $T=4$. These initial results establish the feasibility of sparse latent thought for dense spatial reconstruction and motivate broader continuous-thought architectures for dense vision.