---
library_name: diffusers
license: other
license_name: research-use
tags:
  - object-detection
  - bounding-boxes
  - grounding
  - diffusion
  - synthetic-data
---

# socknethd

Higher-resolution joint image + bounding-box diffusion model with a LoRA adapter, a confidence stream, and Nash-MTL multi-task loss weighting.

Released by [Breakpoint AI](https://huggingface.co/BreakpointAI) as part of open-sourcing the company's research
artifacts.

| | |
|---|---|
| Training data | [`BreakpointAI/breakpoint-grounding-55m`](https://huggingface.co/datasets/BreakpointAI/breakpoint-grounding-55m) |
| Checkpoint step | 4,125,000 |
| Training run | [W&B](https://wandb.ai/diffusionexp/train_boxnethd/runs/c60hovpk) |

## Contents

| Path | Size | What it is |
|---|---|---|
| `boxnet/` | 16.1 GB | Joint image + bounding-box diffusion backbone |
| `pytorch_lora_weights.safetensors` | 2.8 GB | LoRA adapter weights |
| `nash_mtl_weights_conf_ema.pkl` | 432 B | Nash-MTL task-weighting coefficients |
| `nash_mtl_weights_ema.pkl` | 424 B | Nash-MTL task-weighting coefficients |

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

## Citation

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