Skip to content

Pipeline Configuration

Jump to bottom
Mohammad Al-Hanoosh edited this page Dec 10, 2025 · 1 revision

Before you begin, you must configure the pipeline using the .env file.

Settings

  • HF_TOKEN: Provide a valid Hugging Face token with approved access to the Segment Anything Model 3 for instance masks: https://huggingface.co/facebook/sam3.

  • DATA_ROOT / OUTPUT_ROOT: These should generally remain unchanged. The entire pipeline is containerized, so you typically have no reason to modify them.

  • TORCH_CUDA_ARCH_LIST: Specify your GPU architecture. Architectures 8.0, 8.6, and 8.9 are already included.

  • CAPTURE_FRAMERATE: The frame rate of the sensor camera during capture. This is only relevant when processing .bag files. You should be capturing at 30 FPS, so this value usually does not need adjustment.

  • SCENE_NAME: The name of the scene you are processing.

  • FPS_EXTRACT: The frame extraction rate used when processing video input.

Wiki

Table of Contents

Clone this wiki locally