The VLM code in generate_vlm.py has paths pointing to a specific local machine which blocks anyone else from running it.
What needs to change:
- Line 5:
DATASET_DIR — hardcoded to /Users/jaimerobles/Desktop/... → use os.getenv("VLM_DATASET_DIR") or a CLI arg
- Line 7:
ASSESSMENTS_DIR — same, hardcoded Mac path
- Line 350:
temp_dir = "/tmp/hurricane_flr_crops" — use tempfile.gettempdir() so it works cross-platform
- Line 39 in generate.py: same hardcoded images path
Once these are configurable, the rest of the team can run the pipeline. The florence dataset is available in the meta repo release for a clean data source.
The VLM code in
generate_vlm.pyhas paths pointing to a specific local machine which blocks anyone else from running it.What needs to change:
DATASET_DIR— hardcoded to/Users/jaimerobles/Desktop/...→ useos.getenv("VLM_DATASET_DIR")or a CLI argASSESSMENTS_DIR— same, hardcoded Mac pathtemp_dir = "/tmp/hurricane_flr_crops"— usetempfile.gettempdir()so it works cross-platformOnce these are configurable, the rest of the team can run the pipeline. The florence dataset is available in the meta repo release for a clean data source.