ASRF + HASR

Computer Vision32 results

• Action Localizationon50 Salads

  Acc

  83.9

  best: 91.4 (Br-Prompt+ASPnet (RGB, flow, accelerometer))
• Action Localizationon50 Salads

  Edit

  81

  best: 89.1 (Semantic2Graph)
• Action Localizationon50 Salads

  F1@10%

  86.6

  best: 92.7 (Br-Prompt+ASPnet (RGB, flow, accelerometer))
• Action Localizationon50 Salads

  F1@25%

  85.7

  best: 91.6 (Br-Prompt+ASPnet (RGB, flow, accelerometer))
• Action Localizationon50 Salads

  F1@50%

  78.5

  best: 88.5 (Br-Prompt+ASPnet (RGB, flow, accelerometer))
• Action LocalizationonGTEA

  Acc

  76.9

  best: 89.8 (Semantic2Graph)
• Action LocalizationonGTEA

  Edit

  84.5

  best: 93.5 (FACT)
• Action LocalizationonGTEA

  F1@10%

  89.2

  best: 96.1 (FACT)
• Action LocalizationonGTEA

  F1@25%

  87.2

  best: 95.6 (FACT)
• Action LocalizationonGTEA

  F1@50%

  74.8

  best: 91.3 (Semantic2Graph)
• Action LocalizationonBreakfast

  Acc

  69.4

  best: 78 (AdaFocus (newly extracted I3D-features, LT-Context model))
• Action LocalizationonBreakfast

  Average F1

  67.1

  best: 76.2 (AdaFocus (newly extracted I3D-features, LT-Context model))
• Action LocalizationonBreakfast

  Edit

  71.9

  best: 79.7 (FACT (efficient hybrid of convolution and transformer model))
• Action LocalizationonBreakfast

  F1@10%

  74.7

  best: 82.1 (AdaFocus (newly extracted I3D-features, LT-Context model))
• Action LocalizationonBreakfast

  F1@25%

  69.5

  best: 79 (AdaFocus (newly extracted I3D-features, LT-Context model))
• Action LocalizationonBreakfast

  F1@50%

  57

  best: 67.5 (AdaFocus (newly extracted I3D-features, LT-Context model))
• Action Segmentationon50 Salads

  Acc

  83.9

  best: 91.4 (Br-Prompt+ASPnet (RGB, flow, accelerometer))
• Action Segmentationon50 Salads

  Edit

  81

  best: 89.1 (Semantic2Graph)
• Action Segmentationon50 Salads

  F1@10%

  86.6

  best: 92.7 (Br-Prompt+ASPnet (RGB, flow, accelerometer))
• Action Segmentationon50 Salads

  F1@25%

  85.7

  best: 91.6 (Br-Prompt+ASPnet (RGB, flow, accelerometer))
• Action Segmentationon50 Salads

  F1@50%

  78.5

  best: 88.5 (Br-Prompt+ASPnet (RGB, flow, accelerometer))
• Action SegmentationonGTEA

  Acc

  76.9

  best: 89.8 (Semantic2Graph)
• Action SegmentationonGTEA

  Edit

  84.5

  best: 93.5 (FACT)
• Action SegmentationonGTEA

  F1@10%

  89.2

  best: 96.1 (FACT)
• Action SegmentationonGTEA

  F1@25%

  87.2

  best: 95.6 (FACT)
• Action SegmentationonGTEA

  F1@50%

  74.8

  best: 91.3 (Semantic2Graph)
• Action SegmentationonBreakfast

  Acc

  69.4

  best: 78 (AdaFocus (newly extracted I3D-features, LT-Context model))
• Action SegmentationonBreakfast

  Average F1

  67.1

  best: 76.2 (AdaFocus (newly extracted I3D-features, LT-Context model))
• Action SegmentationonBreakfast

  Edit

  71.9

  best: 79.7 (FACT (efficient hybrid of convolution and transformer model))
• Action SegmentationonBreakfast

  F1@10%

  74.7

  best: 82.1 (AdaFocus (newly extracted I3D-features, LT-Context model))
• Action SegmentationonBreakfast

  F1@25%

  69.5

  best: 79 (AdaFocus (newly extracted I3D-features, LT-Context model))
• Action SegmentationonBreakfast

  F1@50%

  57

  best: 67.5 (AdaFocus (newly extracted I3D-features, LT-Context model))