Darkroom models
Model files used by Kan's Darkroom, an in-browser photo editor. Photos never leave the browser; these files are downloaded once, on request, and run on the device with onnxruntime-web.
selfie_multiclass_256x256.onnx
Google's MediaPipe multiclass selfie segmenter, converted from TFLite to ONNX. Unchanged apart from the format.
- Original: https://storage.googleapis.com/mediapipe-models/image_segmenter/selfie_multiclass_256x256/float32/latest/selfie_multiclass_256x256.tflite (sha256 c6748b1253a99067ef71f7e26ca71096cd449baefa8f101900ea23016507e0e0)
- Model card: https://storage.googleapis.com/mediapipe-assets/Model%20Card%20Multiclass%20Segmentation.pdf
- Licence: Apache License 2.0, as stated in the model card. Authors: Adel Ahmadyan, Google. A copy of the licence is in LICENSE.
- Changes: converted with tf2onnx 1.17.0 (TensorFlow 2.21, opset 17). Outputs match the TFLite interpreter within 1e-4.
- Input:
input_29, float32 [1, 256, 256, 3], NHWC, RGB scaled to -1..1. - Output:
Identity, float32 [1, 256, 256, 6], scores (apply softmax) for background, hair, body skin, face skin, clothes, others. - sha256 of the ONNX file: 64ba8ecac03f4fbe06af2224c24bffe1e940405aa0a10db1ce54e37d05e0e670
Not intended for surveillance or identity recognition, as the model card says.