Florence-2
A multimodal visual foundation model released by Microsoft that supports a range of vision tasks including object detection, image captioning, and visual question answering. The model incorporates open-vocabulary grounding capabilities across diverse visual understanding benchmarks.
Timeline (last 90 days)
Florence-2 concatenates the feature sequence output by the image vision encoder with the token sequence of the text prompt and feeds it into the encoder, after which the decoder autoregressively generates the text answer, with task differences expressed by the prompt prefix.
The Florence-2 model schema is annotated with 'Grayscale input image', implying that the input may be converted to grayscale, so tasks that depend on color discrimination may not be applicable.
Multimodal models such as Grounding DINO and Florence-2 combine visual encoders with language models to improve localization accuracy under open-vocabulary conditions.
All Facts (3)
Florence-2 concatenates the feature sequence output by the image vision encoder with the token sequence of the text prompt and feeds it into the encoder, after which the decoder autoregressively generates the text answer, with task differences expressed by the prompt prefix.
50%UnverifiedThe Florence-2 model schema is annotated with 'Grayscale input image', implying that the input may be converted to grayscale, so tasks that depend on color discrimination may not be applicable.
50%UnverifiedMultimodal models such as Grounding DINO and Florence-2 combine visual encoders with language models to improve localization accuracy under open-vocabulary conditions.
50%