G. M. Dilshan Godaliyadda

Prateek Chennuri, Yiheng Chi, Enze Jiang et al.

The proliferation of single-photon image sensors has opened the door to a plethora of high-speed and low-light imaging applications. However, data collected by these sensors are often 1-bit or few-bit, and corrupted by noise and strong motion. Conventional video restoration methods are not designed to handle this situation, while specialized quanta burst algorithms have limited performance when the number of input frames is low. In this paper, we introduce Quanta Video Restoration (QUIVER), an end-to-end trainable network built on the core ideas of classical quanta restoration methods, i.e., pre-filtering, flow estimation, fusion, and refinement. We also collect and publish I2-2000FPS, a high-speed video dataset with the highest temporal resolution of 2000 frames-per-second, for training and testing. On simulated and real data, QUIVER outperforms existing quanta restoration methods by a significant margin. Code and dataset available at https://github.com/chennuriprateek/Quanta_Video_Restoration-QUIVER-

Hao Liang, Zhixuan Ge, Ashish Tiwari et al.

We present FastAvatar, a pose-invariant, feed-forward framework that can generate a 3D Gaussian Splatting (3DGS) model from a single face image from an arbitrary pose in near-instant time (<10ms). FastAvatar uses a novel encoder-decoder neural network design to achieve both fast fitting and identity preservation regardless of input pose. First, FastAvatar constructs a 3DGS face ``template'' model from a training dataset of faces with multi-view captures. Second, FastAvatar encodes the input face image into an identity-specific and pose-invariant latent embedding, and decodes this embedding to predict residuals to the structural and appearance parameters of each Gaussian in the template 3DGS model. By only inferring residuals in a feed-forward fashion, model inference is fast and robust. FastAvatar significantly outperforms existing feed-forward face 3DGS methods (e.g., GAGAvatar) in reconstruction quality, and runs 1000x faster than per-face optimization methods (e.g., FlashAvatar, GaussianAvatars and GASP). In addition, FastAvatar's novel latent space design supports real-time identity interpolation and attribute editing which is not possible with any existing feed-forward 3DGS face generation framework. FastAvatar's combination of excellent reconstruction quality and speed expands the scope of 3DGS for photorealistic avatar applications in consumer and interactive systems.

Yan Zhang, G. M. Dilshan Godaliyadda, Nicola Ferrier et al.

Analytical electron microscopy and spectroscopy of biological specimens, polymers, and other beam sensitive materials has been a challenging area due to irradiation damage. There is a pressing need to develop novel imaging and spectroscopic imaging methods that will minimize such sample damage as well as reduce the data acquisition time. The latter is useful for high-throughput analysis of materials structure and chemistry. In this work, we present a novel machine learning based method for dynamic sparse sampling of EDS data using a scanning electron microscope. Our method, based on the supervised learning approach for dynamic sampling algorithm and neural networks based classification of EDS data, allows a dramatic reduction in the total sampling of up to 90%, while maintaining the fidelity of the reconstructed elemental maps and spectroscopic data. We believe this approach will enable imaging and elemental mapping of materials that would otherwise be inaccessible to these analysis techniques.

G. M. Dilshan P. Godaliyadda, Dong Hye Ye, Michael D. Uchic et al.

Sparse sampling schemes have the potential to dramatically reduce image acquisition time while simultaneously reducing radiation damage to samples. However, for a sparse sampling scheme to be useful it is important that we are able to reconstruct the underlying object with sufficient clarity using the sparse measurements. In dynamic sampling, each new measurement location is selected based on information obtained from previous measurements. Therefore, dynamic sampling schemes have the potential to dramatically reduce the number of measurements needed for high fidelity reconstructions. However, most existing dynamic sampling methods for point-wise measurement acquisition tend to be computationally expensive and are therefore too slow for practical applications. In this paper, we present a framework for dynamic sampling based on machine learning techniques, which we call a supervised learning approach for dynamic sampling (SLADS). In each step of SLADS, the objective is to find the pixel that maximizes the expected reduction in distortion (ERD) given previous measurements. SLADS is fast because we use a simple regression function to compute the ERD, and it is accurate because the regression function is trained using data sets that are representative of the specific application. In addition, we introduce a method to terminate dynamic sampling at a desired level of distortion, and we extended the SLADS methodology to sample groups of pixels at each step. Finally, we present results on computationally-generated synthetic data and experimentally-collected data to demonstrate a dramatic improvement over state-of-the-art static sampling methods.