Advancements in Synthetic Video Generation for Autonomous Driving
Figure
4-2 : Multi SEAN residual block .
The inputs to these SEAN blocks are Flow , Image , and Label embeddings based on encoderdecoder architecture . Label encoding uses an encoder-decoder style 20 to embed input labels to distinctive features , which acts as one of the inputs to Multi SEAN in the Image generator . Flow embedding deals with optical flow outputs of the previous frame and is again fed through the Multi SEAN block of the image generator . Image and segmentation generator encodes image and segmented frames , respectively . Image encoder uses previously generated frames while segmentation encoder uses first frame semantics .
Figure 4-3 describes our system ' s Flow or Label embedding , Image , and Segmentation decoder . This helps to generate inputs for the image generator .
20 https :// ieeexplore . ieee . org / document / 9076374 120
March 2024