--------------------------------------------------
Semantic video analysis evaluation framework
--------------------------------------------------
Purpose:
Evaluation of semantic video analysis approaches to: a) motion-based recognition, b) modality fusion and temporal context exploitation

Task:
Semantic shot classification

Developer:
Georgios Th. Papadopoulos (papad@iti.gr)


--------------------------------------------------
Datasets
--------------------------------------------------
Tennis
Total number of annotated shots: 1615
Supported classes: rally, serve, replay, break

News
Total number of annotated shots: 1188
Supported classes: anchor, reporting, reportage, graphics

Volleyball-I
Total number of annotated shots: 1356
Supported classes: rally, serve, replay, break

Volleyball-II
Total number of annotated shots: 1412
Supported classes: rally, ace, serve, serve preparation, replay, player celebration, tracking single player, face close-up, tracking multiple players


--------------------------------------------------
Available data
--------------------------------------------------
Ground truth annotations:
	-annotated_videos_training1.txt: annotated videos for training the motion-based classifier
	-annotated_videos_training2.txt: annotated videos for training the multimodal-based (fusion and temporal context) classifier (single-modality analysis results are available)
	-annotated_videos_test.txt: annotated videos for evaluating the motion-based and the multimodal classifier (single-modality analysis results are available)

Estimated motion fields (using optical flow estimation algorithm):
	-motion_videoFileName-frameNumber_horizontal.txt: horizontal motion component of frame 'frameNumber' of video 'videoFileName' (positive values denote rightward motion)
	-motion_videoFileName-frameNumber_vertical.txt: vertical motion component of frame 'frameNumber' of video 'videoFileName' (positive values denote downward motion)
	
Single-modality analysis results (a posterior probability is given for every possible shot-class combination, which denotes the degree to which each shot is associated with every semantic class):
	-audio_analysis.txt: Available audio-based analysis results
	-color_analysis.txt: Available color-based analysis results
	-motion_analysis.txt: Available motion-based analysis results