Guide person re-identification workflows with BNNeck trick and loss configuration.
npx skills add https://github.com/AxGord/claude-workflow --skill domain-reid
Please help me install this Agent Skill. Skill: domain-reid Source: https://github.com/AxGord/claude-workflow/tree/main/templates/skills/domain-reid Command: npx skills add https://github.com/AxGord/claude-workflow --skill domain-reid
This Skill unit provides critical insights and gotchas in person re-identification, ensuring that machine learning workflows in this domain are optimized and avoid common pitfalls.
Review the 'BNNeck Trick' to understand distance metric direction for effective person re-identification.
High-intent search queries and answers about installing and using this skill.
The BNNeck trick for person re-identification separates feature spaces by applying batch normalization before the identity classifier, allowing Euclidean distance for triplet loss training and cosine distance for inference.
The best way to balance loss in person re-identification is combining cross-entropy, triplet mining, and center loss. This mixture ensures training stability by simultaneously optimizing classification boundaries, inter-class distance, and intra-class feature compactness.
CLIP-ReID works through a two-stage process for person re-identification. It first optimizes text prompts to learn visual-language prompts, then fine-tunes the image encoder using the learned prompts to achieve state-of-the-art performance.
Standard datasets for benchmarking person re-identification performance provide specific characteristics for evaluating models. Comparing your results against these state-of-the-art performance numbers helps validate that your workflow avoids common ML pitfalls.
Triplet loss training fails on person re-identification tasks when distance metric directions are misapplied. Using the wrong distance metric during feature processing disrupts the separation of intra-class and inter-class feature distributions.
thinking-machines-labConfigure and run Direct Preference Optimization training with a model and preference dataset.
floflo777Evaluate RAG systems using retrieval, generation, and latency metrics.
tachyon-beepRoute training problems to optimization specialists based on symptoms.
umputunRe-analyze stalled problems and generate 2-3 alternative approaches with trade-offs.
