google-research/frame-interpolation 🖼️🔢 → 🖼️

▶️ 278.1K runs 📅 Feb 2022 ⚙️ Cog 0.4.1 🔗 GitHub 📄 Paper ⚖️ License

image-to-video video-frame-interpolation

About

Frame Interpolation for Large Scene Motion

Example Output

Output

[object Object]

Performance Metrics

50.00s Prediction Time

114.97s Total Time

All Input Parameters

{
  "frame1": "https://replicate.delivery/mgxm/5de85319-a354-4178-a2b0-aab4a65fa480/start.png",
  "frame2": "https://replicate.delivery/mgxm/aebabf54-c730-4efe-857d-1182960918d4/end.png",
  "times_to_interpolate": "7"
}

Input Parameters

frame1 (required) Type: string: The first input frame
frame2 (required) Type: string: The second input frame
times_to_interpolate Type: integerDefault: 1Range: 1 - 8: Controls the number of times the frame interpolator is invoked If set to 1, the output will be the sub-frame at t=0.5; when set to > 1, the output will be the interpolation video with (2^times_to_interpolate + 1) frames, fps of 30.

Output Schema

Output

Type: string • Format: uri

Example Execution Logs

2022-02-13 17:20:45.241662: I tensorflow/stream_executor/cuda/cuda_dnn.cc:368] Loaded cuDNN version 8101
Interpolated frames generated, saving now as output video.

Version Details

Version ID: 4f88a16a13673a8b589c18866e540556170a5bcb2ccdc12de556e800e9456d3d
Version Created: September 30, 2022

Run on Replicate →