Resources

Dataset · Code

Dataset

• GraphDroid Benchmark — ASE 2026
  A benchmark for mobile app GUI testing, comprising 41 open-source and commercial Android apps across 21 categories. The reproduction package accompanies our work on history-aware exploration and intent fulfillment.
  Project Page

• VP-Bench — AAAI 2026
  A benchmark for visual prompting in multimodal large language models. Its perception stage contains 34,267 images and 38,932 questions, while a second stage evaluates the effect of visual prompts on six downstream tasks, including GUI grounding and medical imaging.
  Project Page

• KG-RAG UI Transition Graphs — EMNLP 2025
  A collection of UI transition graphs for GUI agent research, covering 30 Chinese apps on Android and HarmonyOS, plus 12 English apps on Android. The graphs support exploration and retrieval of app interaction paths for intent-driven navigation.
  Project Page

• ICM-Instruct Validation Sets — AAAI 2025
  Released validation data for rule-based, explainable image content moderation. The collection includes 1,623 images across 35 categories and a 379-image subset for evaluating moderation explanation quality.
  Project Page

• SpaceSGG — WACV 2025
  An instruction-tuning dataset for scene graph generation and spatial reasoning, combining object locations, relations, and depth information. It provides spatial descriptions, question-answer pairs, and multi-turn conversations, together with a 271-question spatial relation validation set.
  Project Page

• QR Dataset — IEEE TIP 2024
  A dataset for joint denoising and deblurring of dual-exposure Quad-Bayer images, pairing RAW inputs with RGB ground truth. It contains 701 synthetic training tuples, 30 validation tuples, and real captures from an IMX586 sensor.
  Project Page

• D2-Dataset — ECCV 2022
  A dataset for night image restoration from long- and short-exposure image pairs, introduced with D2HNet. It provides 5,661 synthetic training tuples, validation data, and 28 real captured image pairs for studying joint denoising and deblurring.
  Project Page

• Family Face Database (FF-Database) — IEEE TMM 2022
  A dataset for child face prediction from parent images, introduced with ChildPredictor. It contains 7,488 parent faces and 8,558 child faces with labeled facial attributes. Access is available by request under the project’s terms of use.
  Project Page

• Legacy Photo Dataset (LP Dataset) — WACV 2021
  A collection of approximately 25,000 legacy photographs with real noise and degradation, designed for learning noise priors and restoring old photos. Research access is available by request under the project’s terms of use.
  Project Page

Code

Selected implementations from my first-author papers and a widely used repository I maintain.

• QRNet — IEEE TIP 2024
  Official PyTorch implementation for reconstructing sharp, clean RGB images from dual-exposure Quad-Bayer inputs, with training and evaluation code and pre-trained models.
  Project Page

• SVCNet — IEEE TIP 2023
  Official PyTorch implementation for scribble-guided video colorization, propagating user-provided colors while maintaining temporal consistency across frames.
  Project Page

• HSGAN — IEEE TNNLS 2023
  Official implementation for hyperspectral reconstruction from RGB images using a generative adversarial network, with training and validation code.
  Project Page

• D2HNet — ECCV 2022
  Official PyTorch implementation for joint denoising and deblurring of night photos, combining long- and short-exposure images through a hierarchical restoration network.
  Project Page

• ChildPredictor — IEEE TMM 2022
  Official implementation for child face synthesis conditioned on parent images, using disentangled representations of facial attributes. The repository provides inference code, pre-trained models, and a data collection pipeline.
  Project Page

• VCGAN — IEEE TMM 2022
  Official implementation of a hybrid generative adversarial network for automatic image and video colorization, incorporating temporal information to improve consistency in videos.
  Project Page

• HRNet — CVPR NTIRE Workshop 2020
  Official implementation of a hierarchical regression network for spectral reconstruction from RGB images. The method won first place in the real-world track of the NTIRE 2020 Spectral Reconstruction Challenge.
  Project Page

• SCGAN — IEEE TCSVT 2020
  Official implementation for saliency-guided image colorization, using semantic information and saliency maps to generate plausible colors. Training and evaluation code are provided.
  Project Page

• deepfillv2 — PyTorch reimplementation
  My PyTorch reimplementation of gated-convolution image inpainting, inspired by the ICCV 2019 DeepFill v2 paper by Yu et al. It focuses on gated convolutions in a coarse-to-fine framework and provides training and testing code.
  Project Page