Skip to content

Latest commit

 

History

History
61 lines (53 loc) · 2.24 KB

MOT_README.md

File metadata and controls

61 lines (53 loc) · 2.24 KB
Raw

We currently support fou MOT methods: TransTrack, FairMOT, ByteTrack, and our UTT. More methods could be easily reproduced with our codebase.

Data and Checkpoint structures

data
   |——————coco
   |        └——————train2017
   |        └——————annotations
   └——————GOT10k
   |        └——————train
   |        └——————val
   |        └——————test
   └——————crowdhuman
   |         └——————Crowdhuman_train
   |         └——————Crowdhuman_val
   |         └——————annotation_train.odgt
   |         └——————annotation_val.odgt
   └——————MOT
   |        └——————train
   |        └——————test
checkpoints
   |——————fairmot
   |        └——————dla34.pth
   |——————transtrack
   |        └—————transtrack50.pth
   |——————yolox
   |        └——————yolox.pth
   |——————utt
   |        └——————utt_mot.pth

Test FairMOT

 torchrun --nproc_per_node 8 --master_port 9999 tools/train_dist.py --config-file configs/fairmot/fairmot.yaml --config-func fairmot --mode mot --eval-only

Test TransTrack

torchrun --nproc_per_node 8 --master_port 9999 tools/train_dist.py --config-file configs/transtrack/transtrack.yaml --config-func transtrack --mode mot --eval-only

Test ByteTrack

torchrun --nproc_per_node 8 --master_port 9999 tools/train_dist.py --config-file configs/bytetrack/bytetrack.yaml --config-func bytetrack --mode mot --eval-only

Test UTT

torchrun --nproc_per_node 8 --master_port 9999 tools/train_dist.py --config-file configs/utt/utt.yaml --config-func utt --mode mot --eval-only

For training, preprocessing MOT datasets to generate annotations

python trackron/data/datasets/data_specs/mot_to_coco.py --data-root $MOT_ROOT

preprocessing CrowdHuman datasets to generate annotations

python trackron/data/datasets/data_specs/crowdhuman_to_coco.py --data-root $CRODHUMAN_ROOT