2025-02-10 Camera Calibration

Processing Steps

1. Set up env Create a .env that looks something like this:

export E4E_NAS_CREDS=/home/n.hui.813/e4e-nas-creds

The e4e-nas-creds file should look something like this:

username=nthui
password=abcdef1234

2. Init submodules (git submodule init; git submodule update)
3. Mount raw data to ./data and lens cals to ./lens_cal. (scripts/00_mount.sh)
4. Create ./fishsense-lite/runtime/.max-cpu with the contents being the maximum number of CPUs to utilize.
5. In the ./fishsense-lite/runtime/ directory, execute ./build. Note that this can take around 10 minutes.
6. Interactively run the docker image:

docker run --rm -it --gpus=all -e NVIDIA_DRIVER_CAPABILITIES=all -v ./filtered_raws:/home/ubuntu/mnt:ro -v ./results:/home/ubuntu/Results:rw -v ./configs:/home/ubuntu/Configs:rw -v ./FSL-07D:/home/ubuntu/calibrations:ro -v ./scripts:/home/ubuntu/scripts:ro --shm-size=10.24gb `whoami`/fishsense-lite

See scripts/02_docker_run.sh 7. Prep the docker environment

scripts/03_docker_prep.sh

fsl generate-ray-config --max-cpu 8 --max-gpu 1

8. Preprocess the laser calibrations

scripts/04_preprocess_laser_data.sh

fsl preprocess ./raw_data/ED-00/FSL-09/LaserCalibration/*.ORF --format JPG \
    --lens-calibration ./calibrations/FSL-09D/fsl-09d-lens-raw.pkg \
    --output Results/FSL-09D/processed_lasers/
fsl preprocess ./raw_data/ED-00/FSL-11/LaserCalibration/*.ORF --format JPG \
    --lens-calibration ./calibrations/FSL-11D/fsl-11d-lens-raw.pkg \
    --output Results/FSL-11D/processed_lasers/

9. Copy images to label studio. Note that this should be done outside of docker

scripts/05_copy_label_studio_data.sh 10. Set up https://labeler.e4e.ucsd.edu

Labeling Interface:

<View>
  <KeyPointLabels name="kp-1" toName="img-1">
    <Label value="Red Laser" background="red"/>
    <Label value="Green Laser" background="green"/>
  </KeyPointLabels>
  <Image name="img-1" value="$img" zoom="true" zoomControl="true"/>
</View>

Storage backend:

Storage Type: Synology
Storage Title: 2025-02-10_fs_data
URL: https://e4e-nas.ucsd.edu:6021
Path: /label_studio/2025-02-10_fs_data
Username: label_studio
Password: ********
File Filter Regex: .*JPG
Treat every bucket object as a source file: True

11. Export the label studio project results into ./label_studio_results
12. Split the resulting json using the scripts/06_split_labels.ipynb
13. Run the laser calibration script

scripts/06_laser_checkerboard_calibrate.sh

fsl calibrate-laser ./raw_data/ED-00/FSL-11/LaserCalibration/*.ORF \
    --lens-calibration ./calibrations/FSL-11D/fsl-11d-lens-raw.pkg \
    --rows 14 \
    --columns 10 \
    --square-size 41 \
    --output ./Results/FSL-11D/fsl-11d-laser.pkg \
    -j ./label_studio_results/fsl-11d.json

fsl calibrate-laser ./raw_data/ED-00/FSL-09/LaserCalibration/*.ORF \
    --lens-calibration ./calibrations/FSL-09D/fsl-09d-lens-raw.pkg \
    --rows 14 \
    --columns 10 \
    --square-size 41 \
    --output ./Results/FSL-09D/fsl-09d-laser.pkg \
    -j ./label_studio_results/fsl-09d.json

14. Check results in command line output

Notes

FSL-11D passes with 0.054 RMS

FSL-09D passes with 0.032 RMS