bhky/opennsfw2

Keras implementation of the Yahoo Open-NSFW model

解決的問題

它提供了一個現代且易於使用的 Yahoo Open-NSFW 模型實現，用於檢測圖像和影片中的色情內容，取代已停止支援的原始 Caffe 版本。

工作原理

該專案使用 Keras 3 實作的預訓練 Yahoo 模型。它支援多種後端（TensorFlow 和 JAX），並提供高階 API 來處理單張圖像、圖像批次以及影片畫格。它包含特定的預處理流程（YAHOO 和 SIMPLE），以確保與原始模型權重的相容性。

適用對象

需要將 NSFW 內容檢測整合到應用程式中的開發者，特別是使用 Keras/TensorFlow/JAX 生態系統的開發者。

主要特色

  • 多後端支援：支援 Keras 3（TensorFlow/JAX）和 tf-keras。
  • 影片分析：能夠對影片的每一幀預測 NSFW 機率。
  • Grad-CAM 支援：產生熱力圖，標示出導致預測結果的圖像區域（僅支援 TensorFlow 後端）。
  • 微調功能：提供範例和工具，可使用 TensorFlow 在自訂資料集上訓練或微調模型。

"This project brings the Yahoo Open-NSFW model into the modern Keras ecosystem, making it accessible and easy to use for developers." — @username

相關

  • 專案
    LSH9832/edgeyolo

    EdgeYOLO 是一款無錨點的目標檢測器，針對 Nvidia Jetson 等嵌入式邊緣設備的即時效能進行優化，支援包括 TensorRT 與 RKNN 在內的多種部署格式。

  • 專案
    rgeirhos/Stylized-ImageNet

    一個用於建立 Stylized-ImageNet 的工具，該版本的 ImageNet 會扭曲紋理，以鼓勵 CNN 更優先考慮物體形狀而非紋理，從而提升魯棒性。

  • 專案
    hku-mars/GS-SDF

    一個結合了 Gaussian Splatting 與神經符號距離場 (SDF) 的 LiDAR 增強型系統，旨在實現幾何學上一致的、照片級寫實的渲染與 ................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................thought: 3D surface reconstruction.

  • 專案
    cvignac/DiGress

    使用 Graph Transformer 架構的離散去噪擴散模型，用於生成合成圖和分子結構。