Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
DIPO: Dual-State Images Controlled Articulated Object Generation Powered by Diverse Data
Authors: Ruiqi Wu, Xinjie wang, Liu.Liu, Chun-Le Guo, Jiaxiong Qiu, Chongyi Li, Lichao Huang, Zhizhong Su, Ming-Ming Cheng
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments demonstrate that DIPO significantly outperforms existing baselines in both the resting state and the articulated state, while the proposed PM-X dataset further enhances generalization to diverse and structurally complex articulated objects. |
| Researcher Affiliation | Collaboration | Ruiqi Wu1, 2, 3 Xinjie Wang3 Liu Liu3 Chunle Guo1, 2 Jiaxiong Qiu3 Chongyi Li1, 2 Lichao Huang3 Zhizhong Su3 Ming-Ming Cheng1, 2 1NKIARI, Shenzhen Futian 2VCIP, CS, Nankai University 3Horizon Robotics |
| Pseudocode | Yes | Example Question 1: The following code is a function that generates a layout from a given object in a grid format. def sample_base(grid_x, grid_y, base_size): # generate the base, and return a coordinate list of grids ... def generate_part_in_grid(base, grid_coords, x1, x2, y1, y2,...): # generate coordinates and articulation info of a part ... def generate_layout(info): # convert grid-level layout into coordinates and articulation parameters base, grid_coords = sample_base(...) articulate_tree = [base] for part in info[ part ]: part = generate_part_in_grid(...) articulate_tree.append(part) |
| Open Source Code | Yes | Our code and dataset are available at https://github.com/RQ-Wu/DIPO. |
| Open Datasets | Yes | Our code and dataset are available at https://github.com/RQ-Wu/DIPO. Part Net-Mobility (PM) [48] Articulated Container Dataset (ACD) [14] |
| Dataset Splits | Yes | We follow the dataset split way of SINGAPO [23] to build the training and testing set. Specifically, the training set is made up of 493 articulated objects from the PM [48] dataset, combined with 600 samples from our proposed PM-X dataset. Each object is rendered by BLENDER_EEVEE_NEXT engine to produce dual-state image pairs from 20 random views. We further introduce a complex data augmentation to enhance the performance of the model, which is detailed in the supplementary materials. For evaluation, we use 77 held-out objects from PM, each rendered from two random views, resulting in 144 dual-state test samples. In addition, we include 135 objects from the ACD dataset [14] to further assess the generalizability of the model to out-of-distribution data. |
| Hardware Specification | Yes | All experiments are conducted on 8 NVIDIA 4090 GPUs. |
| Software Dependencies | No | The paper mentions "BLENDER_EEVEE_NEXT engine" for rendering and "Adam W [26]" for optimization, and uses "GPT-4o [3, 1]" and LLMs, but does not provide specific version numbers for any software libraries or development environments used for implementation. |
| Experiment Setup | Yes | We train our model for 200 epochs with a batch size of 20. The model is optimized by Adam W [26] with β = (0.9, 0.99) The learning rate is set to 5 10 4 for the image-conditioned module and 5 10 5 for the base model. |