| { |
| "field_id": "F009", |
| "field": "action recognition", |
| "directions": [ |
| { |
| "direction_id": "G0", |
| "name": "Visual action classification and discrimination", |
| "scope": "Determine action categories from spatial and temporal evidence in video, including general action classification and discrimination between confusable actions or subtle steps." |
| }, |
| { |
| "direction_id": "G1", |
| "name": "Efficient video action recognition", |
| "scope": "Reduce the computational or processing cost of video action recognition while retaining enough temporal evidence for classification." |
| }, |
| { |
| "direction_id": "G2", |
| "name": "Action recognition from structured motion measurements", |
| "scope": "Recognize actions and activities using body motion and activity structure in pose-joint sequences or wearable and ambient sensor time series." |
| }, |
| { |
| "direction_id": "G3", |
| "name": "Transferable action representation learning", |
| "scope": "Learn transferable action representations beyond downstream action labels to reduce recognition’s dependence on task annotations; this includes skeletal motion and more general video pretraining." |
| }, |
| { |
| "direction_id": "G4", |
| "name": "Recognition with complementary or incomplete action evidence", |
| "scope": "Improve recognition through complementary information, guidance, or knowledge transfer between different inputs, and maintain recognition when modalities, views, illumination, or visual detail are insufficient at test time." |
| }, |
| { |
| "direction_id": "G5", |
| "name": "Interaction and collective activity understanding", |
| "scope": "Understand actions through relationships among people, body parts, hands, objects, and scene entities, including individual interactions and group or subgroup activities." |
| }, |
| { |
| "direction_id": "G6", |
| "name": "Transfer to sparsely exemplified or unseen actions", |
| "scope": "Recognize novel action categories with few or no class-specific training examples using semantic descriptions, language knowledge, or a small number of support examples; this also includes localizing temporal instances of novel categories." |
| }, |
| { |
| "direction_id": "G7", |
| "name": "Action recognition across domain shifts", |
| "scope": "Maintain action recognition as scenes, viewpoints, appearances, or sensing conditions change, including adaptation using observations from a specified target domain and generalization to target domains unseen during training." |
| }, |
| { |
| "direction_id": "G8", |
| "name": "Grounding actions in continuous recordings", |
| "scope": "Determine when actions occur in continuous recordings, their successive segments, or which actors perform them. This includes temporal and spatial detection of selected instances, framewise segmentation, and weakly labeled, sparsely labeled, and causal streaming settings." |
| }, |
| { |
| "direction_id": "G9", |
| "name": "Early action inference and anticipation", |
| "scope": "Infer an action’s category from evidence of an action that has not yet fully occurred, or predict a forthcoming action from preceding observations." |
| }, |
| { |
| "direction_id": "G10", |
| "name": "Action-analysis datasets and evaluation", |
| "scope": "Establish researchable and evaluable action-analysis problems centered on annotated resources, task definitions, or systematic comparison protocols." |
| } |
| ] |
| } |
|
|
