Search papers, labs, and topics across Lattice.
This paper introduces Stable-Layers, a novel reinforcement learning framework that fine-tunes a pretrained layer decomposition model using feedback solely from a vision-language model (VLM), eliminating the need for paired supervision. By employing Flow-GRPO with LoRA adaptation and a two-stage evaluation pipeline, the authors effectively address the challenge of unreliable reward signals from VLMs, leading to improved layer separation and reduced reconstruction errors. The results demonstrate that Stable-Layers outperforms the base model on the Crello dataset, showcasing its potential for enhancing image layer decomposition tasks.
Fine-tuning image layer decomposition models without paired supervision leads to significantly improved layer quality and reduced artifacts.
We present Stable-Layers, a reinforcement learning framework that eliminates the need for paired supervision by fine-tuning a pretrained layer decomposition model using only feedback from a vision-language model (VLM). Starting from Qwen-Image-Layered, we apply Flow-GRPO with LoRA adaptation, sampling multiple candidate decompositions per image, scoring them with a VLM, and optimising the policy from group-relative advantages. The key challenge lies in designing a reliable reward signal: VLMs scoring samples in isolation tend to compress their judgements into a narrow band, leaving GRPO with little within-group variance to learn from. We address this with a two-stage evaluation pipeline that pairs structured per-sample scoring across five edit-centric criteria with a grid-based calibration step in which the VLM re-scores all candidates side-by-side. Stable-Layers produces decompositions with stronger layer separation, fewer blank or artifact-heavy layers, and lower per-layer reconstruction error on the Crello dataset compared to the base model.
