✨ [Add] inference/predict or test mode

✏️ [Fix] Process <- Proccess typo

✅ [Update] type hint and mAP calculation func

🌱 [Add] deterministic config for training

🐛 [Fix] display on DDP mode

♻️ [Refactor] dataset, apply tensorlize

🔀 [Merge] remote-tracking branch 'origin/DDP_BUGS' into Lightning

🔀 Merge remote-tracking branch 'origin/DEPLOY' into Lightning

🌐 [Add] logging of optimizre(learning rate)

🐛 [Fix] PIL read image, manually close image

🩹 [Fix] BoxMatcher, change eps and filter function

✨ [Update] back the warm up batch step

🔨 [Update] gradient scaler

🐛 [Fix] bug in DDP sync_dist

🔒️ [Fix] torch version diff in meshgrid

🔊 [Add] supporting of tensorboard

🗃️ [Update] dataloader, remove YOLOLoader allin1

🚑️ [Fix] progress bar in DDP or multiGPU env

🔊 [Update] log item, format value on wandb

🐛 [Update] Momentum of the optimizer

✨ [New] use lightning framework to training!

🐛 [Fix] display emoji bugs, change it to shortcode

🐛 [Fix] Dataloader log error, remove format str

🔨 fix: BoxMatcher.__call__ now returns all zero anchor matched targets and all False valid mask, if input target has zero annotations in it. (#88)

💬 [Update] logging -> loguru, consistency logging

🔨 [FIX] Fixes memory leak (#83)

🔧 [Update] NMS Config if validation

🔀 [Merge] branches 'DEPLOY' and 'DEPLOY' of github.com:WongKinYiu/yolov9mit into DEPLOY

✨ [New] Add a classification head!

🔀 [Merge] branch 'DDP_BUGS' of github.com:WongKinYiu/yolov9mit into DDP_BUGS

🐛 [Fix] Loss scale, scale back with batch_size

🚛 [Move] Tensorlize out of YoloDataset

🐛 [Fix] DDP bugs, sync mAP on diff rank

🐛 [Fix] bbox_nms function (#76)

🐛 [Update] getitem with removing -1 cls padding

⬆️ [Update] dependency and the load cache function

⬆️ [Update] mAP from 0~1 to percentage

🐛 [Add] Tensorlize @ dataloader

🐛 [Fix] mul-GPU @ valid bugs, return mAP @ rk 0

🐛 [Fix] #38 local mAP bugs, one pd match < 1 gt

✨ [Add] mutli-GPU on validation mode

🐛 [Fix] LR schedule bugs, batchidx should increase

🎨 [Fix] logging bugs, only log on rank0 gpu

👽️ [Update] cudaamp, most torch not support amp yet

🐛 [Fix] loss, avg loss if using ddp

🚨 [Update] cuda.amp to amp, fix torch warning

🔀 [Merge] branch 'DATASET' into TRAIN

♻️ [Refactor] the dataloader import and default aug

🦺 [Add] a RemoveOutlier handle outside bbox

✨ [Add] a RandomCrop transform, align paper