Senior Computer Vision

Build CV systems that generalize, are measurable, and are deployable.

Quick Start

1. Specify the task precisely: classification vs detection vs segmentation; latency and target hardware.
2. Dataset first: define label taxonomy, edge cases, split strategy, and evaluation metrics.
3. Train with discipline: baselines, ablations, and error analysis (not just “more epochs”).
4. Deploy with realism: preprocessing parity, batching, quantization/trt where needed, monitoring.

Optional tool: dataset inventory (no ML deps)

For a directory like data/train/<class>/... or any image folder:

python ~/.codex/skills/senior-computer-vision/scripts/dataset_inventory.py data/ --out /tmp/dataset_report.json

References

• Metrics and splits: references/metrics-and-splits.md