tao-train-reid

To train a person re-identification model, you use metric learning to build discriminative embeddings for matching individuals across different camera views. This Skill supports training with AutoML and manual model tuning to optimize retrieval metrics.

Person re-identification models are evaluated using retrieval-based metrics like CMC and mAP. These metrics measure the accuracy of matching individuals across different camera feeds during the evaluation phase.

Yes, running TAO person re-identification tasks requires Docker and the nvidia-container-toolkit to be installed in your environment. These dependencies are necessary to execute the training, evaluation, and inference workflows.

Yes, you can export trained re-identification models to the ONNX format. This allows you to generate deployment-ready models for running inference on new person re-identification datasets.

Metric learning for cross-camera person matching trains models to generate discriminative embeddings that cluster images of the same person together. It optimizes retrieval metrics like CMC and mAP to accurately identify individuals across multiple camera feeds.

Yes, you can automate model tuning for person re-identification training using the built-in AutoML support. This allows you to optimize retrieval metrics like CMC and mAP without fully manual hyperparameter configuration.