Search

Choreography is a type of art in which movement is designed. Because it involves many complex movements, a lot of time is spent creating the choreography. In this study, we constructed a dataset of dance movements with more than 700,000 frames using dance videos of the “Odotte-mita” genre. Our dataset could have been very useful for recent research in motion generation, as there has been a lack of dance datasets. Furthermore, we verified the choreography of the dance generated by the deep learning method (acRNN) by physically repeating the dance steps. To the best of our knowledge, this is the first time such a verification has been attempted. It became clear that the choreography generated by machine learning had unique movements and rhythms for dancers. In addition, dancing machine-learning-generated choreography provided an opportunity to make new discoveries. In addition, the fact that the model was a stick figure made the choreography vague, and, although it was difficult to remember, it created varied dance steps.

This research investigated the capabilities of Cycle-Consistent adversarial network (Cycle GAN) in the application of stick figure sketches to cartoon character translation. Few studies focused on the task of generating a variety of poses and facial expression of cartoon characters from simple sketches of stick figures, based on unpaired dataset samples. Furthermore, existing studies showed low performance in detecting rare pose features. In this research, two datasets have been created which consists of paired and unpaired images of manually drawn sketches and cartoon characters. The performance of Cycle GAN has been compared against a paired based model, Pix2Pix, by using qualitative and quantitative evaluation measurements. Results show that Pix2Pix outperforms Cycle GAN in accurately mapping cartoon characters to stick figures. Despite that, the Cycle GAN still managed to produce competing results.

As an important part of animation production, the existing method for drawing and rendering the CG animated characters according to motion information mostly relays on expensive manual processing. By adopting the conditional adversarial learning, an automatic animation rendering system for geometry structure attribute is proposed, using a deep convolution generative adversarial network called “pix2pixHD”. A training database containing a variety of motion stick figure is established for different virtual characters to achieve an end-to-end training system to verify this idea. The trained generator is the desired CG animation creator which shows great performance on visual quality and time efficiency proved by the experimental results.

Bharatnatyam is an ancient Indian Classical Dance form consisting of complex postures and movements. One main challenge which has not been addressed till now in the intelligent systems community is to perform pose recognition for the basic postures of this dance form called the Bhangas and use this for expertise prediction. In this paper, pose recognition is performed for some important postures in Bharatnatyam in order to find the origin of these postures from the Bhangas and further use this result to predict the expertise of a Bharatnatyam dancer. The features extracted are 10 joint angles using 15 joint locations to predict the 22 postures derived from the basic postures (Bhangas). Support Vector Machine classifier with a radial basis function kernel performed the best for pose recognition. By performing stick figure analysis and grouping of labels we estimate the origin of each of these postures from the Bhangas. This is followed by verification of the grouping using Hamming distance calculation. Testing is done on our own Bharatnatyam dataset consisting of 102 dancers, achieving an accuracy of 87.14%. Expertise prediction of the dancers for the 22 poses was performed for four ratings - Excellent, Good, Satisfactory and Poor giving an accuracy of 68.46% without grouping of postures and 80.80% with grouping of postures.

This work aims at designing a fuzzy matching algorithm that would automatically recognize an unknown ballet posture from seventeen fundamental ballet dance primitives. A novel and simple 7-stage system is proposed to achieve the desired objective. Minimized skeletons of the dance postures are generated after performing skin color segmentation on them. Straight line approximation on the minimized skeletons with the help of chain code and sampling generate their equivalent stick figure diagrams. Significant straight lines from the stick figure diagrams are considered, their fuzzy membership with respect to the 4 quadrants are evaluated. Finally with the help of the evaluated data, a fuzzy T-norm operator determines the proximity of a generated dance posture with the seventeen fundamental dance primitives.