Add ONNX model inference implementations for MoGe-2, BEN2, and DA3

- Introduced `onnx_moge2.py` for MoGe-2 ONNX inference with NumPy post-processing.
- Added `onnx_ben2.py` for BEN2 ONNX inference with Pillow and NumPy preprocessing.
- Created `onnx_da3.py` for Depth Anything 3 inference with ONNX Runtime and NumPy.
- Implemented shared ONNX Runtime session helpers in `onnx_runtime.py`.
- Developed unit tests for server functionalities in `test_server.py`, including model caching and job management.
- Enhanced job server to handle frame extraction without requiring models.
This commit is contained in:
2026-08-17 10:37:55 +08:00
parent b3a07cde32
commit c5b77c5c85
58 changed files with 5840 additions and 9931 deletions

View File

@@ -18,7 +18,7 @@
参数会集中显示在独立边框中。`Remove Background` 与 `Unshaded` 直接显示,不再使用 Options 折叠;
`Keep Original` 默认关闭,并在分割线下方保持为最后一项。仅当开启 `Remove Background` 时才需要
`BEN2 Base`,并把 `.foreground` 的 Alpha 作为独立
Mask,不在 Service 中改写深度值。保存 `.foreground` 时保留 BEN2 Alpha,但 RGB 使用未经修改的原始输入;
Mask,不在 Server 中改写深度值。保存 `.foreground` 时保留 BEN2 Alpha,但 RGB 使用未经修改的原始输入;
因此材质直接用 `.foreground` 同时提供颜色和 Mask,不再额外加载原图 Image Texture。
Image Empty 引用视频或序列帧时,DA3 与 MoGe-2 会按 Empty 的 Image Frames 数值逐帧估计
深度(未设置时为完整帧数,显式设为 1 时只处理 1 帧),起始帧跟随 Empty 的 Offset,与背景蒙版范围一致;