yolo-world

_posts/computer_vision/papers/2024-02-21-yolo-world.md

@@ -31,4 +31,5 @@ tag: [yolo]
 - `During inference, the text encoder can be removed and the text embeddings can be re-parameterized into weights of RepVL-PAN for efficient deployment.`
 - `open-vocabulary pre-training scheme for YOLO detectors through region-text contrastive learning on largescale datasets, which unifies detection data, grounding data, and image-text data into region-text pairs`
 
-- `prompt-then-detect paradigm`: `encodes the prompts of a user to build an offline vocabulary and the vocabulary varies with different needs. Then, the efficient detector can infer the offline vocabulary on the fly without re-encoding the prompts. For practical applications, once we have trained the detector, i.e., YOLO-World, we can pre-encode the prompts or categories to build an offline vocabulary and then seamlessly integrate it into the detector.`
+- `prompt-then-detect paradigm`: `encodes the prompts of a user to build an offline vocabulary and the vocabulary varies with different needs. Then, the efficient detector can infer the offline vocabulary on the fly without re-encoding the prompts. For practical applications, once we have trained the detector, i.e., YOLO-World, we can pre-encode the prompts or categories to build an offline vocabulary and then seamlessly integrate it into the detector.`
+