AI 용어집
인공지능 완전 사전
VideoGAN
GAN architecture specialized in generating short video sequences by exploiting spatio-temporal structures to maintain consistency between consecutive frames.
Spatio-temporal Discriminator
Critical component of Video GANs that simultaneously evaluates the spatial coherence of each frame and the temporal coherence between consecutive frames.
Temporal Coherence Loss
Additional loss function in Video GANs that penalizes temporal inconsistencies by measuring differences between successive frames or optical flow differences.
Progressive Video Growing
Training technique where the GAN progressively generates videos of increasing resolution, stabilizing learning and improving the final quality of generated sequences.
Action-Conditioned Video Generation
Approach where GANs generate videos conditioned by specified actions or movements, allowing precise control of the generated dynamic content.
3D Convolutional GAN
Architecture using 3D convolutions to directly process the spatio-temporal volume (height × width × time), naturally capturing temporal relationships in videos.
Temporal Motion Attention
Attention mechanism that focuses on relevant regions of temporal motion, improving the consistency and quality of transitions between frames in video generation.
Frame Prediction Network
Component that predicts future frames based on past frames, essential for maintaining narrative and visual consistency in long generated video sequences.
Optical Flow GAN
Specialized GAN that explicitly generates and uses optical flow to guide the generation of consecutive frames, ensuring realistic and natural movements in videos.
SV2V
Sketch-to-Video Translation using GANs to convert animated drawings or sketches into complete and smooth animated video sequences.
Temporal Pyramid Network
Architecture capturing temporal dependencies at multiple time scales, improving consistency for both fast movements and slow transitions in generated videos.
Video Interpolation GAN
GAN specialized in creating intermediate frames between two existing frames, allowing to slow down or speed up video sequences while maintaining smooth motion.