Academia.eduAcademia.edu

Figure 7 – uploaded by Kranthi Kumar

See full PDFdownloadDownload figure
Figure 7: Bar plots showing scores provided by users in response to the four questions (Q1-Q4) posed for (left) all clips, (middle) theatre clips and (right) movie clips. Error bars denote unit standard deviation.

Figure 7 Bar plots showing scores provided by users in response to the four questions (Q1-Q4) posed for (left) all clips, (middle) theatre clips and (right) movie clips. Error bars denote unit standard deviation.

Related Figures (7)

Figure 1: We present an algorithm to retarget a widescreen recording to smaller aspect ratios. The original recording with overlaid eye gaze data from multiple users (each viewer is a unique color) and the results computed by our algorithm (white cropping window) are shown. Our algorithm is content agnostic and can be used to edit a theatre recording from a static camera (top) or re-edit a movie scene (bottom).
Figure 1: We present an algorithm to retarget a widescreen recording to smaller aspect ratios. The original recording with overlaid eye gaze data from multiple users (each viewer is a unique color) and the results computed by our algorithm (white cropping window) are shown. Our algorithm is content agnostic and can be used to edit a theatre recording from a static camera (top) or re-edit a movie scene (bottom).
Figure 2: The recorded eye gaze data from five users and the corresponding output (the optimal x position) of our algorithm. Results with three different type of gaze have been presented i.e. (a) fixation: when the user maintains the eye gaze around a location, our algorithm outputs a static segment; (b) pursuit: a gradual movement of eye gaze results in a pan; (c) saccade: an abrupt transition in gaze leads to cut.
Figure 2: The recorded eye gaze data from five users and the corresponding output (the optimal x position) of our algorithm. Results with three different type of gaze have been presented i.e. (a) fixation: when the user maintains the eye gaze around a location, our algorithm outputs a static segment; (b) pursuit: a gradual movement of eye gaze results in a pan; (c) saccade: an abrupt transition in gaze leads to cut.
K. Rachavarapu, M. Kumar, V. Gandhi & R. Subramanian / Watch to Edit Figure 3: The x-coordinate of the recorded gaze data of 5 users for the sequence "Death of a Salesman" (top row). Gaussian filtered gaze matrix over all users, used as an input for optimization algorithm (middle row). The output of optimization: The optimal state sequence (black line) along with the detected cuts (black circles) for an output aspect ratio of 1:1(bottom row).
K. Rachavarapu, M. Kumar, V. Gandhi & R. Subramanian / Watch to Edit Figure 3: The x-coordinate of the recorded gaze data of 5 users for the sequence "Death of a Salesman" (top row). Gaussian filtered gaze matrix over all users, used as an input for optimization algorithm (middle row). The output of optimization: The optimal state sequence (black line) along with the detected cuts (black circles) for an output aspect ratio of 1:1(bottom row).
Figure 4: An example sequence where our method performs zoom-in and zooms-out action. The original sequence with overlaid gaze date (Top). The output of our algorithm without zoom (Middle) and with zoom (Bottom).
Figure 4: An example sequence where our method performs zoom-in and zooms-out action. The original sequence with overlaid gaze date (Top). The output of our algorithm without zoom (Middle) and with zoom (Bottom).
Table 2: Comparing our method with GDR based on the mean percentage of gaze data included within the retargeted video at 4:3 aspect ratio. Table 1: Parameters for path optimization algorithm and cut de- tection algorithm.
Table 2: Comparing our method with GDR based on the mean percentage of gaze data included within the retargeted video at 4:3 aspect ratio. Table 1: Parameters for path optimization algorithm and cut de- tection algorithm.
Figure 5: The figure shows original frames and overlaid outputs from our method and GDR (coloured, within white rectangles) on a long sequence. Plot shows the x-position of the center of the cropping windows for our method (black curve) and GDR (red curve) over time. Gaze data of 5 users for the sequence are overlaid on the plot. Unlike GDR, our method does not involve hard assumptions and is able to bette include gaze data (best viewed under zoom).
Figure 5: The figure shows original frames and overlaid outputs from our method and GDR (coloured, within white rectangles) on a long sequence. Plot shows the x-position of the center of the cropping windows for our method (black curve) and GDR (red curve) over time. Gaze data of 5 users for the sequence are overlaid on the plot. Unlike GDR, our method does not involve hard assumptions and is able to bette include gaze data (best viewed under zoom).
Figure 6: Frames from the original sequence cropped by the out- put of our algorithm and GDR (white rectangles). Corresponding gaze data (below) reveals that gaze is broadly cluttered around the shown character. Our algorithm produces a perfectly static virtual camera segment (black curve), while GDR results in unmotivated camera movement (red curve). Of the 20 snippets, 14 were extracted from movie scenes, and the remaining from theatre recordings. Since the content of these scenes varied significantly (professionally edited vs wide an-
Figure 6: Frames from the original sequence cropped by the out- put of our algorithm and GDR (white rectangles). Corresponding gaze data (below) reveals that gaze is broadly cluttered around the shown character. Our algorithm produces a perfectly static virtual camera segment (black curve), while GDR results in unmotivated camera movement (red curve). Of the 20 snippets, 14 were extracted from movie scenes, and the remaining from theatre recordings. Since the content of these scenes varied significantly (professionally edited vs wide an-
Related topics:
Computer ScienceArtificial IntelligenceComputer Vision
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved