boxnethd

Higher-resolution joint image + bounding-box diffusion model with a LoRA adapter. Includes EMA weights.

Released by Breakpoint AI as part of open-sourcing the company's research artifacts.

Training data	BreakpointAI/breakpoint-grounding-55m
Checkpoint step	475,000
Training run	W&B

Contents

Path	Size	What it is
ema/	8.6 GB	EMA weights for the backbone
boxnet/	7.4 GB	Joint image + bounding-box diffusion backbone
pytorch_lora_weights.safetensors	1.2 GB	LoRA adapter weights

Inference weights only. Optimizer, LR scheduler, RNG and dataloader state were not uploaded, so this checkpoint cannot be used to resume training.

Citation

@misc{breakpoint_boxnethd,
  title  = {boxnethd},
  author = {Wang, Franklin and Zhong, Desmond and Murdoch, Jamie},
  year   = {2026},
  url    = {https://huggingface.co/BreakpointAI/boxnethd}
}