Osprey icon indicating copy to clipboard operation
Osprey copied to clipboard

[CVPR2024] The code for "Osprey: Pixel Understanding with Visual Instruction Tuning"

Results 13 Osprey issues
Sort by recently updated
recently updated
newest added

Some weights of the model checkpoint at checkpoints/osprey_7b were not used when initializing OspreyLlamaForCausalLM: ['model.vision_tower.vision_tower.visual.trunk.stages.2.blocks.15.weight', 'model.vision_tower.vision_tower.visual.trunk.stages.0.blocks.2.weight', 'model.vision_tower.vision_tower.visual.trunk.stages.2.blocks.12.weight', 'model.vision_tower.vision_tower.visual.trunk.stages.2.blocks.10.weight', 'model.vision_tower.vision_tower.visual.trunk.stages.2.blocks.23.weight', 'model.vision_tower.vision_tower.visual.trunk.stages.2.blocks.18.weight', 'model.vision_tower.vision_tower.visual.trunk.stages.2.blocks.16.weight', 'model.vision_tower.vision_tower.visual.trunk.stages.2.blocks.4.weight', 'model.vision_tower.vision_tower.visual.trunk.stages.2.blocks.7.weight', 'model.vision_tower.vision_tower.visual.trunk.stages.0.blocks.0.weight', 'model.vision_tower.vision_tower.visual.trunk.stages.2.blocks.1.weight', 'model.vision_tower.vision_tower.visual.trunk.stages.1.blocks.2.weight', 'model.vision_tower.vision_tower.visual.trunk.stages.2.blocks.22.weight', 'model.vision_tower.vision_tower.visual.trunk.stages.1.blocks.1.weight', 'model.vision_tower.vision_tower.visual.trunk.stages.2.blocks.17.weight', 'model.vision_tower.vision_tower.visual.trunk.stages.3.blocks.2.weight',...

34s !python /content/Osprey/demo/app.py --model sunshine-lwt/Osprey-7b output 2024-01-14 16:21:15.919229: E tensorflow/compiler/xla/stream_executor/cuda/cuda_dnn.cc:9342] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered 2024-01-14 16:21:15.919294:...

Hi, Thank you for sharing your impressive work! I got confused about Table 2: How are the open vocabulary segmentation metrics calculated? Also, could you please explain how Osprey outputs...