---
license: apache-2.0
pipeline_tag: image-segmentation
tags:
- conversational-image-segmentation
- lora
---

# ConverSeg-Net-3B

This repository contains raw checkpoints for **ConverSeg-Net-3B**, introduced in the paper [Conversational Image Segmentation: Grounding Abstract Concepts with Scalable Supervision](https://huggingface.co/papers/2602.13195).

ConverSeg-Net is designed for Conversational Image Segmentation (CIS), which focuses on grounding abstract, intent-driven concepts, including functional and physical reasoning, into pixel-accurate masks.

- **Project Page:** [https://glab-caltech.github.io/converseg/](https://glab-caltech.github.io/converseg/)
- **Code:** [https://github.com/AadSah/ConverSeg](https://github.com/AadSah/ConverSeg)
- **Paper:** [arXiv:2602.13195](https://arxiv.org/abs/2602.13195)

## Important Note

These are **not** Hugging Face `from_pretrained` model files. They are raw checkpoint files and LoRA adapter files meant to be downloaded and used with the official [ConverSeg codebase](https://github.com/AadSah/ConverSeg).

## Download

```bash
git lfs install
git clone https://huggingface.co/aadarsh99/ConverSeg-Net-3B ./checkpoints/ConverSeg-Net-3B
```

## Sample Usage

After cloning the [ConverSeg codebase](https://github.com/AadSah/ConverSeg) and setting up the environment, you can run inference using the `demo.py` script by pointing to the downloaded checkpoint paths:

```bash
python demo.py \
  --final_ckpt ./checkpoints/ConverSeg-Net-3B/ConverSeg-Net_sam2_90000.torch.torch \
  --plm_ckpt ./checkpoints/ConverSeg-Net-3B/ConverSeg-Net_plm_90000.torch \
  --lora_ckpt ./checkpoints/ConverSeg-Net-3B/lora_plm_adapter_90000 \
  --model_cfg sam2_hiera_l.yaml \
  --base_ckpt /path/to/sam2_hiera_large.pt \
  --image /path/to/image.jpg \
  --prompt "the left-most person" \
  --device cuda \
  --out_dir ./demo_outputs
```

## Citation

```bibtex
@misc{sahoo2026conversationalimagesegmentationgrounding,
  title = {Conversational Image Segmentation: Grounding Abstract Concepts with Scalable Supervision},
  author = {Aadarsh Sahoo and Georgia Gkioxari},
  year = {2026},
  eprint = {2602.13195},
  archivePrefix = {arXiv},
  primaryClass = {cs.CV},
  url = {https://arxiv.org/abs/2602.13195}, 
}
```