search

kchengiva / Shift-GCN

The implementation for "Skeleton-Based Action Recognition with Shift Graph Convolutional Network" (CVPR2020 oral).
Other
269 stars 55 forks source link

Trained models #6

Closed vivek87799 closed 4 years ago

vivek87799 commented 4 years ago

Hi, Thank you very much for the wonderful work and a well documented repository. It would be very helpful if you could provide the trained model.

kchengiva commented 4 years ago

Thanks for your interest. We released several trained models in ./save_models/.

vivek87799 commented 4 years ago

Thank you very much for the reply.

vivek87799 commented 4 years ago

Hi @kchengiva

It will be very kind if you could also provide 2 stream trained model. It will be very helpful for my thesis. Sorry for the trouble, it takes days on my system to train.

kchengiva commented 4 years ago

I'm sorry I couldn't be more helpful. We have deleted other saved models on our server.

vivek87799 commented 4 years ago

Thank you for the reply.

Lyndsey-Xing commented 4 years ago

Hello @kchengiva

I write the code like ./config/nturgbd-cross-subject/XXX.yaml to train the NTU 120 dataset, but my results can't achieve the results in the paper, and there is a certain gap.

Could you provide your . /config/XXX.yaml files of NTU120 and Kinetics? Thanks.

kchengiva commented 4 years ago

Hello. The config file of 1 stream NTU120 X-sub is as below. We have not trained on Kinetics.

Experiment_name: ntu120_ShiftGCN_joint_xsub

# feeder
feeder: feeders.feeder.Feeder
train_feeder_args:
  data_path: ./data/ntu120/xsub/train_data_joint.npy
  label_path: ./data/ntu120/xsub/train_label.pkl
  debug: False
  random_choose: False
  random_shift: False
  random_move: False
  window_size: -1
  normalization: False

test_feeder_args:
  data_path: ./data/ntu120/xsub/val_data_joint.npy
  label_path: ./data/ntu120/xsub/val_label.pkl

# model
model: model.shift_gcn.Model
model_args:
  num_class: 120
  num_point: 25
  num_person: 2
  graph: graph.ntu_rgb_d.Graph
  graph_args:
    labeling_mode: 'spatial'

#optim
weight_decay: 0.0001
base_lr: 0.1
step: [60, 80, 100]

# training
device: [0,1,2,3]
only_train_epoch: 1
batch_size: 64
test_batch_size: 64
num_epoch: 140
nesterov: True
Lyndsey-Xing commented 4 years ago

@kchengiva Thanks for your help.