DataEvolver: Let Your Data Build and Improve Itself via Goal-Driven Loop Agents

TL;DR

DataEvolver is a closed-loop system for iterative visual data generation and refinement, enabling goal-driven dataset creation with multiple artifact types and validation mechanisms.

Contribution

It introduces a reusable framework for building visual datasets through explicit goal tracking, review, correction, and acceptance loops, validated on an object-rotation task.

Findings

Our model outperforms unadapted base and multi-angle LoRA models on SpatialEdit.

Ablation studies show improvements from scene-aware generation to feedback correction.

The framework effectively enhances dataset quality through iterative validation and correction.

Abstract

Constructing controllable visual data is a major bottleneck for image editing and multimodal understanding. Useful supervision is rarely produced by a single rendering pass; instead it emerges through iterative generation, inspection, correction, filtering, and export. We present DataEvolver, a closed-loop visual data engine that organizes this process around explicit goals, persistent artifacts, bounded corrective actions, and acceptance decisions. DataEvolver supports multiple artifact types, including RGB images, masks, depth maps, normal maps, meshes, poses, trajectories, and review traces. In the current release, the system operates through two coupled loops: generation-time self-correction within each sample and validation-time self-expansion across dataset rounds. We validate the framework on an image-level object-rotation setting. With a fixed Qwen-Edit LoRA probe, our final Ours+DualGate model outperforms both the unadapted base model and a public multi-angle LoRA on SpatialEdit and a held-out evaluation set. Ablations show a consistent improvement path from scene-aware generation to feedback-driven correction and dual-gated validation. Beyond the released rotation data, our main contribution is a reusable framework for building visual datasets through explicit goal tracking, review, correction, and acceptance loops.