TASK 01
End-to-End Generation
Generate executable UI code from the design image and structured layer representation.
2,861Production designs
97,652Expert annotations
17Commercial platforms
4Requirement categories
Abstract
TaoD2C-Bench evaluates whether multimodal large language models can generate industrial UI code that satisfies implementation requirements—not merely reproduce visual appearance. TaoD2C pairs real production designs with layer-level expert annotations that distinguish required constraints from permitted implementation choices. The benchmark separates end-to-end generation into requirement inference and requirement realization, exposing capability gaps that visual similarity alone cannot reveal.
Benchmark Overview
Three Evaluation Tasks
TASK 02
Requirement Inference
Infer the implementation requirements encoded implicitly in visual and layer-level evidence.
TASK 03
Requirement Realization
Produce code that correctly implements explicitly provided requirements under target-library constraints.