3 个仓库
Pipelines that convert various data types like text, images, and audio into tensors.
Distinct from Text Processing Pipelines: Expands beyond text-only processing to include image and audio feature extraction pipelines.
Explore 3 awesome GitHub repositories matching data & databases · Multimodal Pre-Processing. Refine with filters or upvote what's useful.
Transformers.js is a JavaScript library and web machine learning framework designed to run pretrained transformer models directly in the browser. It serves as a client-side inference engine and a wrapper for the ONNX Runtime, enabling the execution of multimodal AI tasks on user devices without the need for a backend server. The library distinguishes itself by providing a unified toolkit for processing text, image, and audio data locally. This architecture supports privacy-preserving model inference and reduces latency by performing all computations on the client's hardware. Its capabilities
Implements structured tokenizers and feature extractors to convert raw multimodal input into numerical tensors.
Neuraltalk2 是一个深度学习视觉系统,专为自动图像标注而设计。它基于 PyTorch 构建,利用卷积神经网络编码器与循环神经网络解码器相结合的混合架构,从视觉输入生成文本描述。 该项目具有 GPU 加速的训练流水线,能够通过多进程分发将工作负载分配到多个图形处理器上。它支持为静态图像文件和实时视频流生成描述。 该框架包括编码器微调、带有温度控制的束搜索(beam search)文本采样,以及使用行业标准语言指标来评估标注准确性和流畅性的功能。它还提供了用于数据集预处理、模型检查点持久化以及将预测结果导出为结构化 JSON 文件的工具。 该实现以 Jupyter Notebook 的形式提供。
Implements a pipeline to convert raw image and text pairs into tensors for optimized model training.
mmaction2 是一个 PyTorch 视频理解工具箱,专为训练和评估深度学习模型而设计。它作为一个用于动作识别、时间定位和时空动作检测的框架,为基于像素的视频分析和基于骨架的动作识别提供了专业工具。 该项目以其模块化架构而著称,具有基于注册表的组件发现和分层、配置驱动的模型组装功能。它支持多模态特征融合,集成 RGB 帧、光流和音频,并包括用于文本到视频片段检索和零样本视频预测的功能。 该框架广泛涵盖了视频数据集工程,包括标注标准化和帧采样,以及全面的模型训练和评估。它提供用于分布式训练、知识蒸馏和通过模型重参数化进行推理优化的实用程序。 该代码库支持 ONNX 模型导出和用于跨不同计算节点部署的环境容器化。
Extracts and processes 3D and 2D skeletal pose information from video for pose-based recognition.