hustvl/EVF-SAM — Star 增长与排名

Official code of "EVF-SAM: Early Vision-Language Fusion for Text-Prompted Segment Anything Model"

📷 EVF-SAM

Early Vision-Language Fusion for Text-Prompted Segment Anything Model

Yuxuan Zhang1,*, Tianheng Cheng1,*, Lei Liu2, Heng Liu2, Longjin Ran2, Xiaoxin Chen2, Wenyu Liu1, Xinggang Wang1,📧

1 Huazhong University of Science and Technology, 2 vivo AI Lab

(* equal contribution, 📧 corresponding author)

News

2025.3: We've released the latest evaluation benchmark GSEval as the comprehensive and multi-granular grounding benchmark.
2025.1: Preview! the EVF-SAM v2 is on the way, going to support salient object segmentation, salient object matte, and referring matte! Besides, better performance on original capabilities is developed!
2024.9: We have expanded our EVF-SAM to powerful SAM-2. Besides fewer parameters and improvements on image prediction, our new model also performs well on video prediction (powered by SAM-2). Only at the expense of a simple image training process on RES datasets, we find our EVF-SAM has zero-shot video text-prompted capability. Try our code!

Highlight

EVF-SAM extends SAM's capabilities with text-prompted segmentation, achieving high accuracy in Referring Expression Segmentation.
EVF-SAM is designed for efficient computation, enabling rapid inference in few seconds per image on a T4 GPU.

Updates

[x] Release code
[x] Release weights
[x] Release demo 👉 🤗 evf-sam
[x] Release code and weights based on SAM-2
[x] Update demo supporting SAM-2👉 🤗 evf-sam2
[x] release new checkpoint supporting body part segmentation and semantic level segmentation.
[x] update demo supporting multitask

Visualization

Input text

Input image

Output

"zebra top left"

"a pizza with a yellow sign on top of it"

"the broccoli closest to the ketchup bottle"

"[semantic] hair"

"[semantic] sea"

Installation

Clone this repository
Install pytorch for your cuda version. Note that torch>=2.0.0 is needed if you are to use SAM-2, and torch>=2.2 is needed if you want to enable flash-attention. (We use torch==2.0.1 with CUDA 11.7 and it works fine.)
pip install -r requirements.txt
If you are to use the video prediction function, run:

cd model/segment_anything_2
python setup.py build_ext --inplace

Weights

Name

SAM

BEIT-3

Params

Prompt Encoder & Mask Decoder

Reference Score

EVF-SAM-multitask

SAM-H

BEIT-3-L

1.32B

train

84.2

EVF-SAM2-multitask

SAM-2-L

BEIT-3-L

898M

freeze

83.2

EVF-SAM

SAM-H

BEIT-3-L

1.32B

train

83.7

EVF-SAM2

SAM-2-L

BEIT-3-L

898M

freeze

83.6

EVF-Effi-SAM-L

EfficientSAM-S

BEIT-3-L

700M

train

83.5

EVF-Effi-SAM-B

EfficientSAM-T

BEIT-3-B

232M

train

80.0

-multimask checkpoints are only available with commits>=9d00853, while other checkpoints are available with commits<9d00853
-multimask checkpoints are jointly trained on Ref, ADE20k, Object365, PartImageNet, humanparsing, pascal part datasets. These checkpoints are able to segment part (e.g., hair, arm), background object (e.g., sky, ground), and semantic-level masks. (by adding special token "[semantic] " in front your prompt)

Inference

1. image prediction

python inference.py  \
  --version  \
  --precision='fp16' \
  --vis_save_path "" \
  --model_type <"ori" or "effi" or "sam2", depending on your loaded ckpt>   \
  --image_path  \
  --prompt

--load_in_8bit and --load_in_4bit are optional
for example:

python inference.py  \
  --version YxZhang/evf-sam2 \
  --precision='fp16' \
  --vis_save_path "vis" \
  --model_type sam2   \
  --image_path "assets/zebra.jpg" \
  --prompt "zebra top left"

2. video prediction

firstly slice video into frames

ffmpeg -i .mp4 -q:v 2 -start_number 0 /'%05d.jpg'

then:

python inference_video.py  \
  --version  \
  --precision='fp16' \
  --vis_save_path "vis/" \
  --image_path    \
  --prompt    \
  --model_type sam2

you can use frame2video.py to concat the predicted frames to a video.

Demo

image demo

python demo.py

video demo

python demo_video.py

Data preparation

Referring segmentation datasets: refCOCO, refCOCO+, refCOCOg, refCLEF (saiapr_tc-12) and COCO2014train

├── dataset
│   ├── refer_seg
│   │   ├── images
│   │   |   ├── saiapr_tc-12 
│   │   |   └── mscoco
│   │   |       └── images
│   │   |           └── train2014
│   │   ├── refclef
│   │   ├── refcoco
│   │   ├── refcoco+
│   │   └── refcocog

Evaluation

torchrun --standalone --nproc_per_node  eval.py   \
    --version  \
    --dataset_dir    \
    --val_dataset "refcoco|unc|val" \
    --model_type <"ori" or "effi" or "sam2", depending on your loaded ckpt>

Acknowledgement

We borrow some codes from LISA, unilm, SAM, EfficientSAM, SAM-2.

Citation

@article{zhang2024evfsamearlyvisionlanguagefusion,
      title={EVF-SAM: Early Vision-Language Fusion for Text-Prompted Segment Anything Model}, 
      author={Yuxuan Zhang and Tianheng Cheng and Rui Hu and Lei Liu and Heng Liu and Longjin Ran and Xiaoxin Chen and Wenyu Liu and Xinggang Wang},
      year={2024},
      eprint={2406.20076},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2406.20076}, 
}