Słownik AI

📖

pojęcia

Visual Self-Attention

Mechanism allowing each image patch to evaluate its relative importance with respect to all other patches to capture global dependencies without convolution.

📖

pojęcia

Cross-Attention Detection

Bidirectional mechanism where object queries interact with image features to simultaneously localize and classify objects.

📖

pojęcia

Token-to-Token ViT

Variant introducing a progressive transition between tokens with resizing and recombination to preserve local structural information.

📖

pojęcia

Transformer Decoder Head

Final module of DETR architectures transforming encoder features into bounding box predictions and classes via attention on object queries.