| --- |
| license: mit |
| pipeline_tag: image-to-image |
| library_name: onnxruntime |
| tags: |
| - onnx |
| - rife |
| - frame-interpolation |
| - video |
| - motion-interpolation |
| --- |
| |
| # RIFE_FP32 |
| |
| RIFE_FP32 is an ONNX export of a RIFE frame interpolation model intended for video frame generation and FPS upscaling workflows. |
|
|
| ## Model Description |
|
|
| This repository provides a float32 ONNX version of a RIFE model for frame interpolation. |
| It is intended for use in applications that generate intermediate frames between two input video frames in order to increase perceived smoothness or raise output framerate. |
|
|
| ## File |
|
|
| - `RIFE_fp32.onnx` |
|
|
| ## Intended Use |
|
|
| This model is intended for: |
|
|
| - video frame interpolation |
| - FPS upscaling workflows |
| - offline processing pipelines |
| - ONNX Runtime based applications |
|
|
| Example use cases include: |
|
|
| - converting 24 FPS video to higher framerates |
| - generating in-between frames for smoother playback |
| - integrating frame interpolation into custom desktop tools or video pipelines |
|
|
| ## Input / Output |
|
|
| The model is expected to take paired frame data prepared by the calling application and output interpolated intermediate frames. |
|
|
| Exact tensor formatting, preprocessing, and batching may depend on the application using the model. |
|
|
| ## Usage Notes |
|
|
| - Designed for ONNX Runtime inference |
| - Best suited for integration into custom frame interpolation pipelines |
| - Performance depends on hardware, ONNX Runtime provider, and input resolution |
| - CUDA, DirectML, ROCm, or CPU execution may be used depending on the environment |
|
|
| ## Limitations |
|
|
| - Quality may vary on fast motion, occlusion boundaries, transparency, particles, or scene cuts |
| - Results depend heavily on preprocessing and postprocessing in the host application |
| - This repository provides the model file only, not a full standalone interpolation application |
|
|
| ## License |
|
|
| This model repository is released under the MIT License. |
|
|
| ## Acknowledgments |
|
|
| RIFE is widely used for video frame interpolation research and practical workflows. |
| This repository provides an ONNX-packaged version for deployment and application integration. |
