Constrained Generative Modeling with Manually Bridged Diffusion Models
Authors: Saeid Naderiparizi, Xiaoxuan Liang, Berend Zwartsenberg, Frank Wood
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | 5 Experiments We demonstrate MBM on a simple 2D synthetic dataset and a traffic scenario generation experiment with collision and offroad avoidance. Additionally, we include an image watermarking experiment in the appendix. We release the source code implementing MBM together with the 2D synthetic and image watermarking experiments. |
| Researcher Affiliation | Collaboration | 1Department of Computer Science, University of British Columbia, Vancouver, Canada 2Alberta Machine Intelligence Institute (Amii), Edmonton, Canada 3Inverted AI, Vancouver, Canada |
| Pseudocode | No | The paper describes the methodology using mathematical formulations and textual explanations but does not include any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | Code and extended version github.com/plai-group/manually-bridged-models. We release the source code implementing MBM together with the 2D synthetic and image watermarking experiments. |
| Open Datasets | No | The paper describes the use of a "simple 2D synthetic dataset" and data from a "traffic scenario generation experiment" but does not provide any specific links, DOIs, repositories, or formal citations for public access to these datasets. |
| Dataset Splits | No | Our dataset consists of 1,000 samples from this data distribution (Fig. 3b). The problem in this experiment is to generate up to 25 vehicles on a bird s-eye view image of a road from one of 70 locations in the dataset. The paper mentions dataset sizes and origins but does not provide specific details on how the data was split into training, validation, or test sets. |
| Hardware Specification | No | This research was enabled in part by technical support and computational resources provided by the Digital Research Alliance of Canada Compute Canada (alliancecan.ca), the Advanced Research Computing at the University of British Columbia (arc.ubc.ca), and Amazon. The paper mentions general computing resources but does not specify particular hardware components such as GPU or CPU models, or memory specifications. |
| Software Dependencies | No | The paper does not provide specific ancillary software details, such as library names with version numbers (e.g., Python, PyTorch, CUDA versions) needed to replicate the experiment. |
| Experiment Setup | No | The γ functions for collision and offroad are respectively γc(t) = 1 / (10σ2(t)) and γo(t) = 1 / (100σ2(t)). It takes the DB-arch models around 250k iterations to achieve their maximum validation ELBO while the other models achieve the maximum at around 20k-30k iterations. While specific iteration counts and bridge function definitions are given, the paper lacks other explicit details on experimental setup such as learning rates, batch sizes, optimizers, or model initialization strategies. |