Skip to content

SAM-HQ

Segment Anything in High Quality (SAM-HQ) works like SAM and consists of pretrained models that perform image segmentation on RGB images and can aid the human in the annotation process.

Prerequisites#

Linux#

  • docker
  • redis-server (sudo apt install redis-server)

Windows#

  • WSL2 using 22.04.x

    • docker (instructions)
    • redis-server (sudo apt install redis-server)

Directories#

sam-hq
|
+-- cache    # cache directory for Pytorch-related files 
|
+-- models   # for storing the SAM-HQ models

You can create the structure using the following command:

mkdir -p sam-hq/cache \
mkdir -p sam-hq/models

Pretrained models#

Pretrained models can be downloaded from here, with the medium-sized vit_l being the recommended one (requires <6GB GPU RAM). vit_l is used in the commands below.

From within the sam-hq/models directory, run the following command:

wget https://huggingface.co/lkeab/hq-sam/resolve/main/sam_hq_vit_l.pth

Service scripts (WSL2 without Docker Desktop UI)#

  • Create a bash script happy_samhq_start.sh in /usr/local/bin with the following content:
#!/bin/bash
redis-server &
dockerd &
seq 10 | xargs -I{} sh -c "echo waiting...; sleep 1;"

  • Make the script executable with sudo chmod a+x happy_samhq_start.sh

  • Create a bash script happy_samhq_stop.sh in /usr/local/bin with the following content:

#!/bin/bash
killall redis-server
killall dockerd
  • Make the script executable with sudo chmod a+x happyhq_sam_stop.sh

SAM scripts#

In the sam-hq directory, create script start.sh with the following content:

#!/bin/bash

scriptdir=`dirname -- "$0";`

docker run --pull always --rm \
  -u $(id -u):$(id -g) -e USER=$USER \
  -v $scriptdir/cache:/.cache \
  -v $scriptdir:/workspace \
  --gpus=all --net=host \
  -t waikatodatamining/pytorch-sam-hq:2023-08-17_cuda11.6 \
  samhq_predict_redis \
  --redis_in sam_in \
  --redis_out sam_out \
  --model /workspace/models/sam_hq_vit_l.pth \
  --model_type vit_l \
  --verbose

And make executable with chmod a+x start.sh.

NB: This script uses the sam_in and sam_out Redis channels to make it a drop-in replacement for SAM in the happy-envi-viewer.

Next, create a script called stop.sh with the following content:

#!/bin/bash

ids=`ps a | grep [s]amhq_predict_redis | sed s/"^[ ]*"//g | cut -f1 -d" "`
for id in $ids
do
  kill -9 $id
done

And make executable with chmod a+x stop.sh.

Starting#

Docker and Redis (WSL2 without Docker Desktop UI)#

sudo /usr/local/bin/happy_samhq_start.sh

Wait till the Waiting... output stops, which waits for about 10 seconds after the Docker daemon starts in the background.

SAM#

In the sam-hq directory, execute the start.sh script.

Stopping#

SAM#

In the sam-hq directory, execute the stop.sh script.

Docker and Redis (WSL2 without Docker Desktop UI)#

sudo /usr/local/bin/happy_samhq_stop.sh

NB: This will also stop any running SAM/SAM-HQ process.