Gsmv's picture
|
download
raw
1.26 kB

Image colorizer (browser build)

ONNX export of DDColor-T (piddnad/DDColor, weights piddnad/ddcolor_paper_tiny revision cf9fd99c1d7472689ec7413441c1b799a51866a3, pytorch_model.bin sha256 8a1277bc90a1bfbb6d2d83933a9a6bc821931879ca93e26e4fcec12165d41fce). Licensed under the Apache License 2.0 (see LICENSE); this file is a modified version: exported to ONNX (opset 17) at a fixed 256x256 input with the DDColor code at commit 2adb63f2656ac41cbdf7b894cddd94121a3faf13, simplified with onnxsim, and converted to float16 (onnxruntime convert_float_to_float16, keep_io_types=True), so inputs and outputs stay float32.

Used by the free image colorizer on wavespeed.ai, which runs it in the browser with onnxruntime-web.

Input input: float32 [1, 3, 256, 256], the picture resized to 256x256, reduced to its CIE L channel and converted back to RGB (a = b = 0), values 0-1. Output output: float32 [1, 2, 256, 256], the predicted CIE a and b channels; resize them to the picture's size and combine with the picture's own L channel.

ddcolor-tiny-256-fp16.onnx sha256 c50ff07e9c02285208bac88130dce46c3eb0961464f01e4597fe8f663761f8c7, 113022111 bytes.

Xet Storage Details

Size:
1.26 kB
·
Xet hash:
ef6a09f9da90379103d54df031bdfbc143a626e10fdcde2718987fd70db1cdb9

Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.