Traditional video stabilization methods use a warping operation to smooth the camera path but result in missing regions in the video frames. To solve this issue, full-frame video stabilization techniques attempt to fill in the unidentified boundary regions, but their effectiveness is limited. In this work, we propose a full-frame video stabilization method using spatiotemporal transformers to fill the missing boundary regions after the warping operation. For training, we adopt a self-supervised strategy and improve it by incorporating temporal information. The proposed approach allows the utilization of redundant video information spatially and temporally while filling in missing regions. Experimental results show that our approach achieves superior results on popular video stabilization datasets. The code, pre-trained model, and video results are available at https://github.com/leventkaracan/VidStabFormer.

Increase in popularity of deep convolutional neural networks in many different areas leads to increase in the use of these networks in reinforcement learning. Training a huge deep neural network structure by using simple gradient descent learning can take quite a long time. Some additional learning approaches should be utilized to solve this problem. One of these techniques is use of momentum which accelerates gradient descent learning. Although momentum techniques are mostly developed for supervised learning problems, it can also be used for reinforcement learning problems. However, its efficiency may vary due to the dissimilarities in two training learning processes. In this paper, the performances of different momentum techniques are compared for one of the reinforcement learning problems; Othello game benchmark. Test results show that the Nesterov accelerated momentum technique provided a more effective generalization on benchmark.