File size: 5,151 Bytes
cbe6b2b
 
 
 
 
 
d189598
cbe6b2b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5685102
 
cbe6b2b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b1c06e1
cbe6b2b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
---
license: apache-2.0

language:
  - en
base_model:
  - allenai/Molmo-7B-D-0924
  - Qwen/Qwen2-7B
  - google/siglip-so400m-patch14-384
  - facebook/dinov3-vitl16-pretrain-lvd1689m
pipeline_tag: image-text-to-text
library_name: transformers
tags:
  - multimodal
  - charts
  - diagrams
  - pointing
  - localization
  - CoME-VL
---



<div align="center">
  <h1>CoME-VL: Scaling Complementary Multi-Encoder Vision-Language</h1>
</div>
<p align="center">
  <a href="https://github.com/mbzuai-oryx/CoME-VL">
    <img alt="GitHub" src="https://img.shields.io/badge/GitHub-CoME--VL-black?logo=github">
  </a>
  <a href="https://arxiv.org/abs/2604.03231">
    <img alt="Paper" src="https://img.shields.io/badge/arxiv-2604.03231-blue">
  </a>
  <a href="https://mbzuai-oryx.github.io/CoME-VL/">
    <img alt="Project Page" src="https://img.shields.io/badge/Project-Page-green">
  </a>
  <a href="https://huggingface.co/MBZUAI/CoME-VL">
    <img alt="HuggingFace" src="https://img.shields.io/badge/🤗%20HuggingFace-CoME--VL-yellow">
  </a>
</p>
<div align="center">
  <img src="assets/teaser_fig.png" alt="CoME-VL Teaser" width="800"/>
</div>


## Overview

**CoME-VL** is a complementary multi-encoder vision-language framework that fuses contrastively trained and self-supervised visual representations to improve both visual understanding and grounding. Built on top of [Molmo](https://github.com/allenai/molmo) (Ai2), CoME-VL introduces three key architectural innovations:

- **Entropy-guided layer selection** to identify and select complementary layer ranges from SigLIP2 and DINOv3
- **Orthogonality-regularized multi-layer mixing (OL)** to reduce redundancy and promote complementary feature fusion
- **RoPE-enhanced cross-attention (RGCA)** to spatially align heterogeneous token grids across encoders

<div align="center">
  <img src="assets/main_arct.png" alt="CoME-VL Architecture" width="800"/>
  <p>Overview of CoME-VL: dual encoders (SigLIP2 + DINOv3) fused via orthogonality-regularized mixing and RoPE-based cross-attention, injected into a decoder-only LLM.</p>
</div>

---

## Installation

Python 3.10 is recommended. First install [PyTorch](https://pytorch.org) for your platform, then:

```bash
git clone https://github.com/ankan8145/COME-VL.git
cd COME-VL
pip install -e .[all]
```

---

## Environment Setup

```bash
export MOLMO_DATA_DIR=/path/to/data
export HF_HOME=/path/to/huggingface/cache
```

---

## Training / Fine-tuning

Fine-tune starting from a pretrained checkpoint:

```bash
HF_HUB_OFFLINE=1 \
TRANSFORMERS_OFFLINE=1 \
WANDB_MODE=offline \
WANDB_API_KEY="<your_wandb_key>" \
WANDB_PROJECT="come-vl" \
WANDB_ENTITY="<your_entity>" \
CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 \
torchrun --standalone --nnodes=1 --nproc_per_node=8 \
  launch_scripts/train_multitask_model.py \
  3.2-synthetic \
  checkpoint_folder \
  --save_folder=output_folder \
  --save_overwrite
```

**Notes:**
- `checkpoint_folder` should point to your starting model checkpoint directory.
- `--save_folder` should use a short, descriptive name — avoid long paths with special characters.
- `3.2-synthetic` specifies the training data mixture.
- `--save_overwrite` allows overwriting an existing save folder.

---

## Evaluation

```bash
torchrun --nproc-per-node 1 --master_port 29504 \
  launch_scripts/eval_downstream.py \
  checkpoint_folder \
  "test-low-res" \
  --save_to_checkpoint_dir
```

**Notes:**
- `test-low-res` evaluates at standard resolution on the test split.
- Use `test-high-res` for high-resolution evaluation (add `--fsdp --high_res` flags).
- Results and predictions are saved into the checkpoint directory.
- Add `--overwrite` to re-run and replace cached metrics.

---

## Model Architecture

CoME-VL uses:

- **Language backbone:** Qwen2-7B
- **Contrastive encoder:** SigLIP2-SO400M — semantic alignment
- **Self-supervised encoder:** DINOv3-Large — spatial grounding
- **Selected layers:** SigLIP2 layers 0–27 (all) + DINOv3 layers 10–23 (entropy-guided)

---

## Data

Most data is managed via HuggingFace Datasets. Training uses the [PixMo dataset](https://huggingface.co/collections/allenai/pixmo-674746ea613028006285687b) and RefCOCO.

Download all datasets:

```bash
python3 scripts/download.py all --n_proc 12
```

Download a specific dataset:

```bash
python3 scripts/download_data.py pixmo_count_counting --n_proc 12
```

---

## Pretrained Model Initialization

Convert HuggingFace weights before training from scratch:

```bash
python3 scripts/convert_hf_to_molmo.py qwen2_7b
python3 scripts/convert_hf_to_molmo.py openai
```

---
---

## Citation

If you find CoME-VL useful in your research, please consider citing:
```bibtex
@article{comevl2026,
  title={CoME-VL: Scaling Complementary Multi-Encoder Vision-Language},
  author={Deria, Ankan and Kumar, Komal and He, Xilin and Razzak, Imran and Cholakkal, Hisham and Khan, Fahad Shahbaz and Khan, Salman},
  journal={arXiv preprint},
  year={2026}
}
```

---

## Acknowledgements

This codebase is built on top of **[Molmo](https://github.com/allenai/molmo)** by the Allen Institute for AI (Ai2). We thank the Ai2 team for open-sourcing their work.