Abstract: Techniques are provided for global (non-rigid) scan point registration between a scanned object and an associated model, from an arbitrary initial starting position, based on a combination of iterative coarse registration and fine registration. A methodology implementing the techniques according to an embodiment includes generating a model transformation based on a coarse registration between the model and the point scan. The method further includes calculating an alignment metric based on a distance measurement between the point scan and the transformed model. If the alignment metric exceeds a selected threshold value, a fine registration is performed between the transformed model and the point scan. Otherwise, the method continues by performing a random rotation of the model; a translation of the rotated model towards a centroid of the point scan; and iterating the coarse registration using the translated model until the alignment metric is achieved, after which the fine registration is performed.

Abstract: A system and method for random sampled convolutions are disclosed to efficiently boost a convolutional neural network (CNN) expressive power without adding computation cost. The method for random sampled convolutions selects a receptive field size and generates filters with a subset of the receptive field elements, the number of learnable parameters, as being active, wherein the number learnable parameters corresponds to computing characteristics, such as SIMD capability, of the processing system upon which the CNN is executed. Several random filters may be generated, with each being run separately on the CNN. The random filter that causes the fastest convergence is selected over the others. The placement of the random filter in the CNN may be per layer, per channel, or per convergence operation. The CNN employing the random sampled convolutions method performs as well as other CNNs utilizing the same receptive field size.

Abstract: Techniques are provided for global (non-rigid) scan point registration between a scanned object and an associated model, from an arbitrary initial starting position, based on a combination of iterative coarse registration and fine registration. A methodology implementing the techniques according to an embodiment includes generating a model transformation based on a coarse registration between the model and the point scan. The method further includes calculating an alignment metric based on a distance measurement between the point scan and the transformed model. If the alignment metric exceeds a selected threshold value, a fine registration is performed between the transformed model and the point scan. Otherwise, the method continues by performing a random rotation of the model; a translation of the rotated model towards a centroid of the point scan; and iterating the coarse registration using the translated model until the alignment metric is achieved, after which the fine registration is performed.

Abstract: A method, a computer and a non-transitory computer readable medium for deblurring, the method may include receiving an input image; calculating, based on the input image, a first estimated blur kernel; calculating a first estimate of a latent image based on the input image and the first estimated blur kernel; and performing at least one repetitions of: receiving a current estimate of the latent image; calculating, based on the current estimate of the latent image, a next estimated blur kernel; and calculating a next estimate of the latent image based on the current estimate of the latent image and the next estimated blur kernel