Efficient Grounded-SAM

We're going to combine Grounding-DINO with efficient SAM variants for faster annotating.

Table of Contents

Installation
Efficient SAM Series
Run Grounded-FastSAM Demo
Run Grounded-MobileSAM Demo
Run Grounded-LightHQSAM Demo
Run Grounded-Efficient-SAM Demo
Run Grounded-Edge-SAM Demo
Run Grounded-RepViT-SAM Demo

Installation

Install Grounded-SAM
Install Fast-SAM
Note that we may use the sam image as the demo image in order to compare the inference results of different efficient-sam variants.

Efficient SAMs

Here's the list of Efficient SAM variants:

Title	Description	Links
FastSAM	The Fast Segment Anything Model(FastSAM) is a CNN Segment Anything Model trained by only 2% of the SA-1B dataset published by SAM authors. The FastSAM achieve a comparable performance with the SAM method at 50× higher run-time speed.	[Github] [Demo]
MobileSAM	MobileSAM performs on par with the original SAM (at least visually) and keeps exactly the same pipeline as the original SAM except for a change on the image encoder. Specifically, we replace the original heavyweight ViT-H encoder (632M) with a much smaller Tiny-ViT (5M). On a single GPU, MobileSAM runs around 12ms per image: 8ms on the image encoder and 4ms on the mask decoder.	[Github]
Light-HQSAM	Light HQ-SAM is based on the tiny vit image encoder provided by MobileSAM. We design a learnable High-Quality Output Token, which is injected into SAM's mask decoder and is responsible for predicting the high-quality mask. Instead of only applying it on mask-decoder features, we first fuse them with ViT features for improved mask details. Refer to Light HQ-SAM vs. MobileSAM for more details.	[Github]
Efficient-SAM	Segment Anything Model (SAM) has emerged as a powerful tool for numerous vision applications. However, the huge computation cost of SAM model has limited its applications to wider real-world applications. To address this limitation, we propose EfficientSAMs, light-weight SAM models that exhibit decent performance with largely reduced complexity. Our idea is based on leveraging masked image pretraining, SAMI, which learns to reconstruct features from SAM image encoder for effective visual representation learning. Further, we take SAMI-pretrained light-weight image encoders and mask decoder to build EfficientSAMs, and finetune the models on SA-1B for segment anything task. Refer to EfficientSAM arXiv for more details.	[Github]
Edge-SAM	EdgeSAM involves distilling the original ViT-based SAM image encoder into a purely CNN-based architecture, better suited for edge devices. We carefully benchmark various distillation strategies and demonstrate that task-agnostic encoder distillation fails to capture the full knowledge embodied in SAM. Refer to Edge-SAM arXiv for more details.	[Github]
RepViT-SAM	Recently, RepViT achieves the state-of-the-art performance and latency trade-off on mobile devices by incorporating efficient architectural designs of ViTs into CNNs. Here, to achieve real-time segmenting anything on mobile devices, following MobileSAM, we replace the heavyweight image encoder in SAM with RepViT model, ending up with the RepViT-SAM model. Extensive experiments show that RepViT-SAM can enjoy significantly better zero-shot transfer capability than MobileSAM, along with nearly 10× faster inference speed. Refer to RepViT-SAM arXiv for more details.	[Github]

Run Grounded-FastSAM Demo

Firstly, download the pretrained Fast-SAM weight here
Run the demo with the following script:

cd Grounded-Segment-Anything

python EfficientSAM/grounded_fast_sam.py --model_path "./FastSAM-x.pt" --img_path "assets/demo4.jpg" --text "the black dog." --output "./output/"

And the results will be saved in ./output/ as:

Input	Text	Output
	"The black dog."

Note: Due to the post process of FastSAM, only one box can be annotated at a time, if there're multiple box prompts, we simply save multiple annotate images to ./output now, which will be modified in the future release.

Run Grounded-MobileSAM Demo

Firstly, download the pretrained MobileSAM weight here
Run the demo with the following script:

cd Grounded-Segment-Anything

python EfficientSAM/grounded_mobile_sam.py --MOBILE_SAM_CHECKPOINT_PATH "./EfficientSAM/mobile_sam.pt" --SOURCE_IMAGE_PATH "./assets/demo2.jpg" --CAPTION "the running dog"

And the result will be saved as ./gronded_mobile_sam_anontated_image.jpg as:

Input	Text	Output
	"the running dog"

Run Grounded-Light-HQSAM Demo

Firstly, download the pretrained Light-HQSAM weight here
Run the demo with the following script:

cd Grounded-Segment-Anything

python EfficientSAM/grounded_light_hqsam.py

And the result will be saved as ./gronded_light_hqsam_anontated_image.jpg as:

Input	Text	Output
	"bench"

Run Grounded-Efficient-SAM Demo

Download the pretrained EfficientSAM checkpoint from here and put it under Grounded-Segment-Anything/EfficientSAM
Run the demo with the following script:

cd Grounded-Segment-Anything

python EfficientSAM/grounded_efficient_sam.py

And the result will be saved as ./gronded_efficient_sam_anontated_image.jpg as:

Input	Text	Output
	"bench"

Run Grounded-Edge-SAM Demo

Download the pretrained Edge-SAM checkpoint follow the official instruction as:

cd Grounded-Segment-Anything
wget -P EfficientSAM/ https://huggingface.co/spaces/chongzhou/EdgeSAM/resolve/main/weights/edge_sam.pth
wget -P EfficientSAM/ https://huggingface.co/spaces/chongzhou/EdgeSAM/resolve/main/weights/edge_sam_3x.pth

Run the demo with the following script:

cd Grounded-Segment-Anything

python EfficientSAM/grounded_edge_sam.py

And the result will be saved as ./gronded_edge_sam_anontated_image.jpg as:

Input	Text	Output
	"bench"

Run Grounded-RepViT-SAM Demo

Download the pretrained RepViT-SAM checkpoint follow the official instruction as:

cd Grounded-Segment-Anything
wget -P EfficientSAM/ https://github.com/THU-MIG/RepViT/releases/download/v1.0/repvit_sam.pt

Run the demo with the following script:

cd Grounded-Segment-Anything

python EfficientSAM/grounded_repvit_sam.py

And the result will be saved as ./gronded_repvit_sam_anontated_image.jpg as:

Input	Text	Output
	"bench"