Hi,
Happy May Day holiday!
I would appreciate your advice on choosing between VL-BEiT and BEiT3. I have an industrial dataset comprising 700,000 images along with corresponding text descriptions. For fault diagnosis, different units should be separated from the original images. I'd like to use BEiT3 for instance segmentation so I can utilize the text information; however, BEiT3 cannot be pretrained on my custom dataset.
Is it possible to use BEiT2’s pretrained weights to replace the image component of BEiT3’s pretrained model weights and then use the resulting BEiT3 model for segmentation? Or should I use BEiT2’s pretrained weights and VL-BEiT for this task?
I would really appreciate it if you could reply!
Hi,
Happy May Day holiday!
I would appreciate your advice on choosing between VL-BEiT and BEiT3. I have an industrial dataset comprising 700,000 images along with corresponding text descriptions. For fault diagnosis, different units should be separated from the original images. I'd like to use BEiT3 for instance segmentation so I can utilize the text information; however, BEiT3 cannot be pretrained on my custom dataset.
Is it possible to use BEiT2’s pretrained weights to replace the image component of BEiT3’s pretrained model weights and then use the resulting BEiT3 model for segmentation? Or should I use BEiT2’s pretrained weights and VL-BEiT for this task?
I would really appreciate it if you could reply!