SAM finetune and inference#34

Open

Mark-ZhouWX wants to merge 47 commits intomindspore-lab:masterfrom

Mark-ZhouWX:master

Commits on Jan 15, 2024

add SAM training and inference model
Mark-ZhouWX
committed
add inference one image
Mark-ZhouWX
committed
decouple image , box and mask in resize and pad transform
Mark-ZhouWX
committed
add lite inference
Mark-ZhouWX
committed
move sam examples to demo
Mark-ZhouWX
committed
move sam from research to official/cv
Mark-ZhouWX
committed
update comment for lite-inference
Mark-ZhouWX
committed
add O2 for eval.py
Mark-ZhouWX
committed
fix bug of maskmiou calculation
Mark-ZhouWX
committed
add sa-1b dataset
Mark-ZhouWX
committed
add model arts config
Mark-ZhouWX
committed
hack work_root
Mark-ZhouWX
committed
inference -> O0
Mark-ZhouWX
committed
add ckpt download link
Mark-ZhouWX
committed
workaround of layernorm2d
Mark-ZhouWX
committed
set drop_over_flow true
Mark-ZhouWX
committed
add text prompt inference(blip2)
Mark-ZhouWX
committed
add text prompt train(blip2)
Mark-ZhouWX
committed
fix bug of sample with no large mask
Mark-ZhouWX
committed
add filter_text_encoder button in saving ckpt
Mark-ZhouWX
committed
adapt model_wrapper to support text and box input
Mark-ZhouWX
committed
add cost time log
Mark-ZhouWX
committed
add text-prompt readme
Mark-ZhouWX
committed
add clip text_encoder
Mark-ZhouWX
committed
1.fixbug of unfreezing text_proj; 2.add dimension projection for blip2
Mark-ZhouWX
committed
change default O2 to O0 to prevent failure of clip training
Mark-ZhouWX
committed
update readme and text-inference script
Mark-ZhouWX
committed
add cloud training config
Mark-ZhouWX
committed
update cloud training dataset config
Mark-ZhouWX
committed
update text-inference
Mark-ZhouWX
committed
add batch inference demos
Mark-ZhouWX
committed
add point finetune
Mark-ZhouWX
committed
add point finetune: remove duplicate graph complie
Mark-ZhouWX
committed
add point finetune: add iter training to config
Mark-ZhouWX
committed
add point finetune: add point inference
Mark-ZhouWX
committed
add point finetune: fix get new point bug, add point transform
Mark-ZhouWX
committed
add point finetune: to amp_level O0
Mark-ZhouWX
committed
add point finetune: add loss scale
Mark-ZhouWX
committed
add point finetune: change from grad sum to mean
Mark-ZhouWX
committed
add point finetune: fix bug of unscale order
Mark-ZhouWX
committed
add point finetune: lr schedule from dynamic to pre-computed
Mark-ZhouWX
committed
add point finetune: prevent duplicate graph compile of grad reducer
Mark-ZhouWX
committed
add point finetune: update config of lr_scheduler
Mark-ZhouWX
committed
add point finetune: fix bug of valid_boxes
Mark-ZhouWX
committed
add point finetune: fix bug of all_finite for distributed training
Mark-ZhouWX
committed
update readme and inference demo
Mark-ZhouWX
committed
update readme of blip2 and clip
Mark-ZhouWX
committed