Video Segmentation on AIDA Seedling Background Corpus

Segmentation results are provided for the video data in the following corpus:
1. LDC2018E53	     AIDA Scenario 1 - Seedling Background Corpus Non Eval V1.0

The data includes the following:
1. A master shot boundary file (.msb).
2. Keyframes for each shot.
3. Thumbnail image files corresponding to the keyframes.

The format of the msb file is as follows:

v_xNieGJPu7Afh2STY     IC0019MUS_1     1   737        0.000000       24.590366
v_xNieGJPu7Afh2STY     IC0019MUS_2   738  1470       24.591000       49.048367
v_xNieGJPu7Afh2STY     IC0019MUS_3  1471  2197       49.049000       73.306366
v_xNieGJPu7Afh2STY     IC0019MUS_4  2198  2925       73.306000       97.597370
v_xNieGJPu7Afh2STY     IC0019MUS_5  2926  3054       97.597000      101.901360
|                       |            |     |          |               |
|                       |            |     |          |               End time of the video shot on that line
|                       |            |     |          Start time of the video shot on that line
|                       |            |     End frame of the video shot on that line
|                       |            Start frame of the video shot on that line
|                       AIDA video shot name (Video_shotID)
Random Video ID generated by cineast


Systems will return the Video_shotID in column 2 of the msb file as the keyframe ID for provenance. The segmentation tool outputs keyframe IDs (and physical png images) assuming 1 keyframe per shot. Systems will have to map the AIDA video element to the random ID in the MSB to locate the folder with this random ID that contains the different keyframes for that video. There is a one-to-one correspondence between the AIDA video and the random ID generated by the tool. 

Note: Videos with dimensions greater than 640 x 480 are scaled down when segmented, and so the keyframes have scaled down dimensions for larger videos. In future segmentation results, this will be adjusted so that the keyframe dimensions match the video dimensions.