Search Results for "stage"

The shap-e repository provides the official code and model release for Shap-E, a conditional generative model designed to produce 3D assets (implicit functions, meshes, neural radiance fields) from text or image prompts. The model is built with a two-stage architecture: first an encoder that maps existing 3D assets into parameterizations of implicit functions, and then a conditional diffusion model trained on those parameterizations to generate new assets. Because it works at the level of implicit functions, Shap-E can render output both as textured meshes and NeRF-style volumetric renderings. ...

...Specifically, we train a neural codec language model (called VALL-E) using discrete codes derived from an off-the-shelf neural audio codec model, and regard TTS as a conditional language modeling task rather than continuous signal regression as in previous work. During the pre-training stage, we scale up the TTS training data to 60K hours of English speech which is hundreds of times larger than existing systems. VALL-E emerges in-context learning capabilities and can be used to synthesize high-quality personalized speech with only a 3-second enrolled recording of an unseen speaker as an acoustic prompt. Experiment results show that VALL-E significantly outperforms the state-of-the-art zero-shot TTS system in terms of speech naturalness and speaker similarity. ...

...Its principal advantage is speed: it can generate 3D assets in just 1–2 minutes on a single GPU, which is significantly faster than many competing text-to-3D models. The model works via a two-stage diffusion approach: first, it uses a text → image diffusion network to produce a synthetic 2D view consistent with the prompt; then a second diffusion model converts that image into a 3D point cloud. While it does not match the fine detail of some slower methods, the tradeoff in speed makes it practical for prototyping and interactive 3D generation. ...

...The problem of mirror symmetry refers to the sudden change of the direction of the bangs near the yaw angle of pi/2. We propose to use an auxiliary discriminator to solve this problem. Note that in the initial stage of training, the auxiliary discriminator must dominate the generator more than the main discriminator does. Otherwise, if the main discriminator dominates the generator, the mirror symmetry problem will still occur. In practice, progressive training is able to guarantee this. We have trained many times from scratch. Adding an auxiliary discriminator stably solves the mirror symmetry problem.

Summarization, translation, sentiment analysis, text-generation and more at blazing speed using a T5 version implemented in ONNX. This package is still in the alpha stage, therefore some functionalities such as beam searches are still in development. The simplest way to get started for generation is to use the default pre-trained version of T5 on ONNX included in the package. Please note that the first time you call get_encoder_decoder_tokenizer, the models are being downloaded which might take a minute or two. ...