Skip to content
Tanja edited this page Jul 17, 2015 · 3 revisions

All dataset are located under /extra/data_sets/UCF-101.

Spatial

/extra/data_sets/UCF-101/frames

All frames have a size of 320x240, except for 4 videos:

  • v_PommelHorse_g05_c01
  • v_PommelHorse_g05_c02
  • v_PommelHorse_g05_c03
  • v_PommelHorse_g05_c04

Those videos have a size of 400x226.

txt-files

cnn: /extra/data_sets/UCF-101/frames/txt-files/cnn

  • train.txt & test.txt: All frames with the corresponding class label are listed. The frames in the file are sorted by video.
  • train_random.txt & test_random.txt: All frames with the corresponding class label are listed. The frames in the file are complety shuffeld.

lstm: /extra/data_sets/UCF-101/frames/txt-files/lstm

  • train.txt & test.txt: 16 frames per video with the corresponding class label are listed. The frames in the file are sorted by video.
  • train_random.txt & test_random.txt: 16 frames per video with the corresponding class label are listed. The frames in the file are complety shuffeld.

databases

Name Used by Location Based on
cnn Spatial CNN /extra/data_sets/UCF-101/lmdb/??? ???
lstm Spatial LSTM, Fusion /extra/data_sets/UCF-101/levelDB/spatial/lstm /extra/data_sets/UCF-101/frames/txt-files/lstm: train.txt and test.txt

Motion

/extra/data_sets/UCF-101/flow

The stack size of all extracted flow frames is 10. So the x- and y-flow for the 10 next frames are used for the flow of a current frame.

All frames have a size of 320x240, except for 4 videos:

  • v_PommelHorse_g05_c01
  • v_PommelHorse_g05_c02
  • v_PommelHorse_g05_c03
  • v_PommelHorse_g05_c04

Those videos have a size of 400x226.

txt-files

cnn_all: /extra/data_sets/UCF-101/flow/txt-files/cnn_all

train.txt & test.txt: All flow frames with the corresponding class label are listed. The flow frames in the file are complety shuffeld. This txt-files are based on the train_random.txt and test_random.txt in /extra/data_sets/UCF-101/frames/txt-files/cnn.

cnn_16_per_video: /extra/data_sets/UCF-101/flow/txt-files/cnn_16_per_video

train.txt & test.txt: 16 flow frames per video with the corresponding class label are listed. The flow frames in the file are complety shuffeld. This txt-files are based on the train_random.txt and test_random.txt in /extra/data_sets/UCF-101/frames/txt-files/lstm.

lstm: /extra/data_sets/UCF-101/flow/txt-files/lstm

train.txt & test.txt: 16 flow frames per video with the corresponding class label are listed. The flow frames in the file are sorted by video. This txt-files are based on the train.txt and test.txt in /extra/data_sets/UCF-101/frames/txt-files/lstm.

databases

Name Used by Location Based on
cnn_all Fudan /extra/data_sets/UCF-101/levelDB/flow/cnn_all /extra/data_sets/UCF-101/flow/txt-files/cnn_all/: train.txt and test.txt
cnn_16_per_video Fudan /extra/data_sets/UCF-101/levelDB/flow/cnn_16_per_video /extra/data_sets/UCF-101/flow/txt-files/cnn_16_per_video: train.txt and test.txt
lstm Motion LSTM, Fusion /extra/data_sets/UCF-101/levelDB/flow/lstm /extra/data_sets/UCF-101/flow/txt-files/lstm: train.txt and test.txt
Clone this wiki locally