-
Notifications
You must be signed in to change notification settings - Fork 1
Home
Welcome to the action-classification wiki!
The home page is for quick reference about general things. We are trying to do work on the ROAD challenge, from ICCV 2021
https://www.youtube.com/watch?v=QQF8zlfEtlI
"I want to run: "
Centernet is located here.
From https://github.com/WATonomous/CenterNet/blob/master/experiments/road.sh
CUDA_VISIBLE_DEVICES=0,1 python ../src/main.py ctdet --exp_id road-${HOSTNAME} --batch_size 10 --lr 5e-4 --gpus 0,1 --num_workers 8 --num_epochs 230 --lr_step 180,210 --dataset road --load_model ../models/ctdet_coco_dla_2x.pth
ctdet_coco_dla_2x.pth is the centernet checkpoint from the original authors pretrained on coco.
You can find this file already on the server at /home/ben/Projects/CenterNet/models/ctdet_coco_dla_2x.pth (in ben's folder)
I believe that the best centernet checkpoint is at
/home/ben/Projects/road-dataset/CenterTrack/exp/tracking/road_exp_delta-ubuntu2_2021-07-29T06:37:33+00:00/model_last.pth
To generate detections for the road dataset, run src/demo.py in the CenterNet repo, passing the above checkpoint. This should write data in the format of /mnt/wato-drive/road/detections/val1_coco.jsonl, which can be used as an input for ACAR evaluation (i.e. the end-to-end model evaluation)
To evaluate the centernet model, run test.py
- Launch ACAR docker,
python ACAR-Net/main.py --config ACAR-Net/configs/ROAD/<config_name>.yaml
write a config with evaluate: False in the config
write a config with evaluate: True in the config
and specify the the model checkpoint in the resume_path field
this basically shows for each action label in how many boxes is it annotated with other action labels in that same box
Stop 80904
[('HazLit', 2140), ('Brake', 13941), ('Wait2X', 9569), ('IncatLft', 1483), ('IncatRht', 2341), ('Xing', 14), ('PushObj', 691), ('Ovtak', 3), ('MovTow', 7), ('MovAway', 2), ('XingFmRht', 3), ('TurLft', 9), ('XingFmLft', 3)]
HazLit 2140
[('Stop', 2140), ('IncatRht', 3)]
MovAway 127014
[('TurRht', 3497), ('Brake', 3858), ('Xing', 2398), ('Ovtak', 257), ('MovRht', 185), ('IncatLft', 1225), ('TurLft', 3025), ('PushObj', 1423), ('IncatRht', 1874), ('MovLft', 99), ('XingFmRht', 800), ('XingFmLft', 258), ('Stop', 2), ('MovTow', 2), ('Rev', 3)]
MovTow 111222
[('IncatLft', 378), ('MovRht', 68), ('TurLft', 509), ('Xing', 2329), ('IncatRht', 860), ('TurRht', 1430), ('XingFmLft', 1496), ('PushObj', 1308), ('Ovtak', 487), ('MovLft', 131), ('Wait2X', 14), ('Brake', 52), ('Stop', 7), ('XingFmRht', 186), ('MovAway', 2)]
Brake 17958
[('Stop', 13941), ('MovAway', 3858), ('IncatLft', 624), ('IncatRht', 1652), ('TurRht', 142), ('TurLft', 99), ('XingFmLft', 16), ('XingFmRht', 91), ('Ovtak', 3), ('MovTow', 52)]
Green 19310
[]
TurRht 5858
[('MovAway', 3497), ('IncatRht', 1337), ('MovTow', 1430), ('XingFmLft', 709), ('XingFmRht', 601), ('Brake', 142), ('IncatLft', 6)]
Wait2X 9695
[('Stop', 9569), ('Mov', 112), ('PushObj', 119), ('MovTow', 14)]
Amber 5808
[]
Red 39406
[]
IncatLft 3163
[('MovTow', 378), ('Stop', 1483), ('Brake', 624), ('MovAway', 1225), ('TurLft', 891), ('TurRht', 6), ('XingFmLft', 3)]
XingFmLft 15687
[('MovTow', 1496), ('TurRht', 709), ('IncatRht', 380), ('PushObj', 302), ('Brake', 16), ('Mov', 117), ('Xing', 3), ('MovAway', 258), ('TurLft', 18), ('IncatLft', 3), ('Stop', 3)]
MovRht 253
[('MovTow', 68), ('MovAway', 185), ('IncatRht', 108), ('Ovtak', 6)]
XingFmRht 10813
[('TurRht', 601), ('Brake', 91), ('MovTow', 186), ('MovAway', 800), ('IncatRht', 82), ('Stop', 3), ('TurLft', 39), ('PushObj', 91)]
Xing 4800
[('MovAway', 2398), ('MovTow', 2329), ('Stop', 14), ('Mov', 58), ('XingFmLft', 3)]
Ovtak 747
[('MovAway', 257), ('MovTow', 487), ('IncatRht', 84), ('MovRht', 6), ('Brake', 3), ('Stop', 3)]
Mov 7782
[('Wait2X', 112), ('XingFmLft', 117), ('PushObj', 103), ('Xing', 58)]
TurLft 3953
[('MovAway', 3025), ('IncatLft', 891), ('MovTow', 509), ('Rev', 36), ('Brake', 99), ('XingFmLft', 18), ('XingFmRht', 39), ('Stop', 9)]
PushObj 3896
[('MovAway', 1423), ('MovTow', 1308), ('Stop', 691), ('Wait2X', 119), ('XingFmLft', 302), ('Mov', 103), ('XingFmRht', 91)]
IncatRht 5392
[('MovAway', 1874), ('Brake', 1652), ('Stop', 2341), ('TurRht', 1337), ('MovTow', 860), ('MovRht', 108), ('Ovtak', 84), ('MovLft', 12), ('XingFmLft', 380), ('XingFmRht', 82), ('HazLit', 3)]
MovLft 230
[('MovTow', 131), ('IncatRht', 12), ('MovAway', 99)]
Rev 39
[('TurLft', 36), ('MovAway', 3)]
['agent_ness'],
# object classes (10 of these)
['Ped', 'Car', 'Cyc', 'Mobike', 'MedVeh', 'LarVeh', 'Bus', 'EmVeh', 'TL', 'OthTL'],
# action classes (19 of these)
['Red', 'Amber', 'Green', 'MovAway', 'MovTow', 'Mov', 'Brake', 'Stop', 'IncatLft', 'IncatRht', 'HazLit', 'TurLft', 'TurRht', 'Ovtak', 'Wait2X', 'XingFmLft', 'XingFmRht', 'Xing', 'PushObj'],
# location classes (12 of these)
['VehLane', 'OutgoLane', 'OutgoCycLane', 'IncomLane', 'IncomCycLane', 'Pav', 'LftPav', 'RhtPav', 'Jun', 'xing', 'BusStop', 'parking'],
# object-action duplexes (39 of these)
['Bus-MovAway', 'Bus-MovTow', 'Bus-Stop', 'Bus-XingFmLft', 'Car-Brake', 'Car-IncatLft', 'Car-IncatRht', 'Car-MovAway', 'Car-MovTow', 'Car-Stop', 'Car-TurLft', 'Car-TurRht', 'Car-XingFmLft', 'Car-XingFmRht', 'Cyc-MovAway', 'Cyc-MovTow', 'Cyc-Stop', 'Cyc-TurLft', 'Cyc-XingFmLft', 'Cyc-XingFmRht', 'LarVeh-Stop', 'MedVeh-IncatLft', 'MedVeh-MovTow', 'MedVeh-Stop', 'MedVeh-TurRht', 'OthTL-Green', 'OthTL-Red', 'Ped-Mov', 'Ped-MovAway', 'Ped-MovTow', 'Ped-PushObj', 'Ped-Stop', 'Ped-Wait2X', 'Ped-Xing', 'Ped-XingFmLft', 'Ped-XingFmRht', 'TL-Amber', 'TL-Green', 'TL-Red'],
# object-action location triplets (68 of these)
['Bus-MovTow-IncomLane', 'Bus-MovTow-Jun', 'Bus-Stop-IncomLane', 'Bus-Stop-VehLane', 'Bus-XingFmLft-Jun', 'Car-Brake-Jun', 'Car-Brake-VehLane', 'Car-IncatLft-Jun', 'Car-IncatLft-VehLane', 'Car-IncatRht-IncomLane', 'Car-IncatRht-Jun', 'Car-MovAway-Jun', 'Car-MovAway-OutgoLane', 'Car-MovAway-VehLane', 'Car-MovTow-IncomLane', 'Car-MovTow-Jun', 'Car-Stop-IncomLane', 'Car-Stop-Jun', 'Car-Stop-VehLane', 'Car-TurLft-Jun', 'Car-TurLft-VehLane', 'Car-TurRht-IncomLane', 'Car-TurRht-Jun', 'Car-XingFmLft-Jun', 'Cyc-MovAway-Jun', 'Cyc-MovAway-LftPav', 'Cyc-MovAway-OutgoCycLane', 'Cyc-MovAway-OutgoLane', 'Cyc-MovAway-VehLane', 'Cyc-MovTow-IncomCycLane', 'Cyc-MovTow-IncomLane', 'Cyc-MovTow-Jun', 'Cyc-MovTow-LftPav', 'Cyc-Stop-IncomCycLane', 'Cyc-Stop-IncomLane', 'Cyc-Stop-Jun', 'Cyc-TurLft-Jun', 'Cyc-XingFmLft-Jun', 'MedVeh-MovTow-IncomLane', 'MedVeh-MovTow-Jun', 'MedVeh-Stop-IncomLane', 'MedVeh-Stop-Jun', 'MedVeh-TurRht-Jun', 'Ped-Mov-Pav', 'Ped-MovAway-LftPav', 'Ped-MovAway-Pav', 'Ped-MovAway-RhtPav', 'Ped-MovTow-IncomLane', 'Ped-MovTow-LftPav', 'Ped-MovTow-RhtPav', 'Ped-MovTow-VehLane', 'Ped-PushObj-LftPav', 'Ped-PushObj-RhtPav', 'Ped-Stop-BusStop', 'Ped-Stop-LftPav', 'Ped-Stop-Pav', 'Ped-Stop-RhtPav', 'Ped-Stop-VehLane', 'Ped-Wait2X-LftPav', 'Ped-Wait2X-RhtPav', 'Ped-XingFmLft-IncomLane', 'Ped-XingFmLft-Jun', 'Ped-XingFmLft-VehLane', 'Ped-XingFmLft-xing', 'Ped-XingFmRht-IncomLane', 'Ped-XingFmRht-Jun', 'Ped-XingFmRht-RhtPav', 'Ped-XingFmRht-VehLane']
['Av-move', 'Av-stop', 'Av-turn-right', 'Av-turn-left', 'Av-overtake', 'Av-move-left','Av-move-right']
Video names and their splits in the road dataset
2014-06-25-16-45-34_stereo_centre_02 ['all', 'train_1', 'train_2', 'val_3']
2014-06-26-09-53-12_stereo_centre_02 ['all', 'val_1', 'train_2', 'train_3']
2014-07-14-14-49-50_stereo_centre_01 ['all', 'train_1', 'train_2', 'train_3']
2014-07-14-15-42-55_stereo_centre_03 ['all', 'train_1', 'train_2', 'train_3']
2014-08-08-13-15-11_stereo_centre_01 ['all', 'train_1', 'train_2', 'train_3']
2014-08-11-10-59-18_stereo_centre_02 ['all', 'train_1', 'train_2', 'train_3']
2014-11-14-16-34-33_stereo_centre_06 ['all', 'train_1', 'val_2', 'train_3']
2014-11-18-13-20-12_stereo_centre_05 ['all', 'train_1', 'train_2', 'val_3']
2014-11-21-16-07-03_stereo_centre_01 ['all', 'train_1', 'val_2', 'train_3']
2014-11-25-09-18-32_stereo_centre_04 ['all', 'val_1', 'train_2', 'train_3']
2014-12-09-13-21-02_stereo_centre_01 ['all', 'train_1', 'train_2', 'train_3']
2015-02-03-08-45-10_stereo_centre_02 ['all', 'train_1', 'train_2', 'train_3']
2015-02-03-19-43-11_stereo_centre_04 ['all', 'train_1', 'val_2', 'train_3']
2015-02-06-13-57-16_stereo_centre_02 ['all', 'train_1', 'train_2', 'train_3']
2015-02-13-09-16-26_stereo_centre_02 ['all', 'val_1', 'train_2', 'train_3']
2015-02-13-09-16-26_stereo_centre_05 ['all', 'train_1', 'train_2', 'train_3']
2015-02-24-12-32-19_stereo_centre_04 ['all', 'train_1', 'train_2', 'train_3']
2015-03-03-11-31-36_stereo_centre_01 ['all', 'train_1', 'train_2', 'val_3']
- The split we use for ACAR is train_1, val_1
- the split we used for Centernet is train_1, val_1