ptv3-industrial-23cls

PTv3 Industrial 23-Class

PTv3 semantic-segmentation checkpoint trained on synthetic Isaac Sim industrial RGB-D/point-cloud data.

The checkpoint metadata records best_metric_value=0.9111804725342209 at epoch 120. This value comes from the training experiment's configured validation metric; reproduce evaluation on the intended dataset before using it as a benchmark.

Files

• ptv3-industrial-23cls.pth: inference-only state_dict
• config.py: merged Pointcept experiment configuration
• sha256sums.txt: integrity hashes

Optimizer, scheduler, and training history were removed from the published checkpoint. The original full checkpoint is therefore not resumable from this file.

Classes

The fixed model order is:

barcode, bracket, cardboard_box, cnc_machine, container, conveyor,
fire_extinguisher, floor, floor_decal, forklift, lamp, pallet,
pallet_trolley, pillar, rack, robot, robot_stand, safety_fence, shelf,
sign, table, wall, workpiece

Reproducible API deployment

The Dockerized FastAPI server, RTX 4070/4080 settings, request contract, operations guide, and checksum verification are documented at:

• https://github.com/mingqian0850/industrial_3d_sem_seg_server/tree/ptv3
• https://github.com/mingqian0850/ptv3/blob/agent/portable-inference-deployment/README_API.md
• Draft PR: https://github.com/mingqian0850/ptv3/pull/1

The deployment pins Pointcept commit d7272257058337969292865e0892b91d2f362b6a.

git clone https://github.com/mingqian0850/ptv3.git
cd ptv3
./deploy_scripts/build_api.sh
GPU_DEVICE=0 MAX_VALID_POINTS=350000 ./deploy_scripts/start_api.sh

Intended use and limitations

This model is intended for research and evaluation on scenes compatible with the synthetic industrial training domain and the published 23-class taxonomy. It may not generalize to unseen sensors, materials, lighting, geometry, or class definitions. Review predictions before any safety-relevant use.