快速目标分割：FastSAM

 # Fast Segment Anything [[`📕Paper`](https://arxiv.org/pdf/2306.12156.pdf)] [[`🤗HuggingFace Demo`](https://huggingface.co/spaces/An-619/FastSAM)] [[`Colab demo`](https://colab.research.google.com/drive/1oX14f6IneGGw612WgVlAiy91UHwFAvr9?usp=sharing)] [[`Replicate demo & API`](https://replicate.com/casia-iva-lab/fastsam)] [~~[`OpenXLab Demo`](https://openxlab.org.cn/apps/detail/zxair/FastSAM)~~] [[`Model Zoo`](#model-checkpoints)] [[`BibTeX`](#citing-fastsam)] [[`Video Demo`](https://youtu.be/yHNPyqazYYU)]  The **Fast Segment Anything Model(FastSAM)** is a CNN Segment Anything Model trained using only 2% of the SA-1B dataset published by SAM authors. FastSAM achieves comparable performance with the SAM method at **50× higher run-time speed**.  **🍇 Updates** - **`2023/11/28`** Recommendation: [Semantic FastSAM](https://github.com/KBH00/Semantic-Fast-SAM), which add the semantic class labels to FastSAM. Thanks to [KBH00](https://github.com/KBH00/Semantic-Fast-SAM) for this valuable contribution. - **`2023/09/11`** Release [Training and Validation Code](https://github.com/CASIA-IVA-Lab/FastSAM/releases). - **`2023/08/17`** Release [OpenXLab Demo](https://openxlab.org.cn/apps/detail/zxair/FastSAM). Thanks to OpenXLab Team for help. - **`2023/07/06`** Added to [Ultralytics (YOLOv8) Model Hub](https://docs.ultralytics.com/models/fast-sam/). Thanks to [Ultralytics](https://github.com/ultralytics/ultralytics) for help 🌹. - **`2023/06/29`** Support [text mode](https://huggingface.co/spaces/An-619/FastSAM) in HuggingFace Space. Thanks a lot to [gaoxinge](https://github.com/gaoxinge) for help 🌹. - **`2023/06/29`** Release [FastSAM_Awesome_TensorRT](https://github.com/ChuRuaNh0/FastSam_Awsome_TensorRT). Thanks a lot to [ChuRuaNh0](https://github.com/ChuRuaNh0) for providing the TensorRT model of FastSAM 🌹. - **`2023/06/26`** Release [FastSAM Replicate Online Demo](https://replicate.com/casia-iva-lab/fastsam). Thanks a lot to [Chenxi](https://chenxwh.github.io/) for providing this nice demo 🌹. - **`2023/06/26`** Support [points mode](https://huggingface.co/spaces/An-619/FastSAM) in HuggingFace Space. Better and faster interaction will come soon! - **`2023/06/24`** Thanks a lot to [Grounding-SAM](https://github.com/IDEA-Research/Grounded-Segment-Anything) for Combining Grounding-DINO with FastSAM in [Grounded-FastSAM](https://github.com/IDEA-Research/Grounded-Segment-Anything/tree/main/EfficientSAM) 🌹. ## Installation Clone the repository locally: ```shell git clone https://github.com/CASIA-IVA-Lab/FastSAM.git ``` Create the conda env. The code requires `python>=3.7`, as well as `pytorch>=1.7` and `torchvision>=0.8`. Please follow the instructions [here](https://pytorch.org/get-started/locally/) to install both PyTorch and TorchVision dependencies. Installing both PyTorch and TorchVision with CUDA support is strongly recommended. ```shell conda create -n FastSAM python=3.9 conda activate FastSAM ``` Install the packages: ```shell cd FastSAM pip install -r requirements.txt ``` Install CLIP: ```shell pip install git+https://github.com/openai/CLIP.git ``` ## <a name="GettingStarted"></a> Getting Started First download a [model checkpoint](#model-checkpoints). Then, you can run the scripts to try the everything mode and three prompt modes. ```shell # Everything mode python Inference.py --model_path ./weights/FastSAM.pt --img_path ./images/dogs.jpg ``` ```shell # Text prompt python Inference.py --model_path ./weights/FastSAM.pt --img_path ./images/dogs.jpg --text_prompt "the yellow dog" ``` ```shell # Box prompt (xywh) python Inference.py --model_path ./weights/FastSAM.pt --img_path ./images/dogs.jpg --box_prompt "[[570,200,230,400]]" ``` ```shell # Points prompt python Inference.py --model_path ./weights/FastSAM.pt --img_path ./images/dogs.jpg --point_prompt "[[520,360],[620,300]]" --point_label "[1,0]" ``` You can use the following code to generate all masks, make mask selection based on prompts, and visualize the results. ```shell from fastsam import FastSAM, FastSAMPrompt model = FastSAM('./weights/FastSAM.pt') IMAGE_PATH = './images/dogs.jpg' DEVICE = 'cpu' everything_results = model(IMAGE_PATH, device=DEVICE, retina_masks=True, imgsz=1024, conf=0.4, iou=0.9,) prompt_process = FastSAMPrompt(IMAGE_PATH, everything_results, device=DEVICE) # everything prompt ann = prompt_process.everything_prompt() # bbox default shape [0,0,0,0] -> [x1,y1,x2,y2] ann = prompt_process.box_prompt(bboxes=[[200, 200, 300, 300]]) # text prompt ann = prompt_process.text_prompt(text='a photo of a dog') # point prompt # points default [[0,0]] [[x1,y1],[x2,y2]] # point_label default [0] [1,0] 0:background, 1:foreground ann = prompt_process.point_prompt(points=[[620, 360]], pointlabel=[1]) prompt_process.plot(annotations=ann,output_path='./output/dog.jpg',) ``` You are also welcomed to try our Colab demo: [FastSAM_example.ipynb](https://colab.research.google.com/drive/1oX14f6IneGGw612WgVlAiy91UHwFAvr9?usp=sharing). ## Different Inference Options We provide various options for different purposes, details are in [MORE_USAGES.md](MORE_USAGES.md). ## Training or Validation Training from scratch or validation: [Training and Validation Code](https://github.com/CASIA-IVA-Lab/FastSAM/releases). ## Web demo ### Gradio demo - We also provide a UI for testing our method that is built with gradio. You can upload a custom image, select the mode and set the parameters, click the segment button, and get a satisfactory segmentation result. Currently, the UI supports interaction with the 'Everything mode' and 'points mode'. We plan to add support for additional modes in the future. Running the following command in a terminal will launch the demo: ``` # Download the pre-trained model in "./weights/FastSAM.pt" python app_gradio.py ``` - This demo is also hosted on [HuggingFace Space](https://huggingface.co/spaces/An-619/FastSAM).   ### Replicate demo - [Replicate demo](https://replicate.com/casia-iva-lab/fastsam) has supported all modes, you can experience points/box/text mode.    ## <a name="Models"></a>Model Checkpoints Two model versions of the model are available with different sizes. Click the links below to download the checkpoint for the corresponding model type. - **`default` or `FastSAM`: [YOLOv8x based Segment Anything Model](https://drive.google.com/file/d/1m1sjY4ihXBU1fZXdQ-Xdj-mDltW-2Rqv/view?usp=sharing) | [Baidu Cloud (pwd: 0000).](https://pan.baidu.com/s/18KzBmOTENjByoWWR17zdiQ?pwd=0000)** - `FastSAM-s`: [YOLOv8s based Segment Anything Model.](https://drive.google.com/file/d/10XmSj6mmpmRb8NhXbtiuO9cTTBwR_9SV/view?usp=sharing) ## Results All result were tested on a single NVIDIA GeForce RTX 3090. ### 1. Inference time Running Speed under Different Point Prompt Numbers(ms). | method | params | 1 | 10 | 100 | E(16x16) | E(32x32\*) | E(64x64) | |:------------------:|:--------:|:-----:|:-----:|:-----:|:----------:|:-----------:|:----------:| | SAM-H | 0.6G | 446 | 464 | 627 | 852 | 2099 | 6972 | | SAM-B | 136M | 110 | 125 | 230 | 432 | 1383 | 5417 | | FastSAM | 68M | 40 |40 | 40 | 40 | 40 | 40 | ### 2. Memory usage | Dataset | Method | GPU Memory (MB) | | :-------: | :-----: | :-------------: | | COCO 2017 | FastSAM | 2608 | | COCO 2017 | SAM-H | 7060 | | COCO 2017 | SAM-B | 4670 | ### 3. Zero-shot Transfer Experiments #### Edge Detection Test on the BSDB500 dataset. |method | year| ODS | OIS | AP | R50 | |:----------:|: