Scope
Input
tank_short_test_3dgs_pack_20frames.zipGate
check_3dgs_training_ready.py must passClaimData plumbing only, not reconstruction quality
Fetch The Pack
mkdir -p /tmp/aqua_3dgs_smoke cd /tmp/aqua_3dgs_smoke curl -L \ -o tank_short_test_3dgs_pack_20frames.zip \ https://github.com/rsasaki0109/aqua_localization/releases/download/v0.3/tank_short_test_3dgs_pack_20frames.zip unzip -o tank_short_test_3dgs_pack_20frames.zip
Check Readiness
ros2 run aqua_localization check_3dgs_training_ready.py \ --pack /tmp/aqua_3dgs_smoke/tank_short_test_3dgs_pack_20frames
3DGS training ready: true format: nerfstudio frames: 20 images: 20 intrinsics source: manual
Training Placeholder
Run this in an external training environment. Keep nerfstudio, gsplat, CUDA, and similar dependencies outside the normal ROS workspace unless you intentionally want them there.
cd /tmp/aqua_3dgs_smoke/tank_short_test_3dgs_pack_20frames # Example shape only. Pin the real tool version in the training PR. ns-train splatfacto \ --data . \ --max-num-iterations 300 \ --output-dir /tmp/aqua_3dgs_smoke/nerfstudio_out
Expected Smoke Outputs
train.logor captured terminal output- first render or viewer screenshot
- training command and tool version
- elapsed time and GPU/CPU note
- failure note if the tool rejects the tiny 20-frame pack
Record the run with the training smoke report template.