Download lapsrn/demo.py from opencv/opencv_contribution: direct link, hf CLI and curl.
- Browser
- Download file 2.4 kB
-
https://huggingface.co/opencv/opencv_contribution/resolve/main/lapsrn/demo.py
- Command line
-
hf download hf://opencv/opencv_contribution/lapsrn/demo.py
-
curl -L -o demo.py https://huggingface.co/opencv/opencv_contribution/resolve/main/lapsrn/demo.py
2.4 kB
| import argparse | |
| import os | |
| import cv2 as cv | |
| import numpy as np | |
| here = os.path.dirname(os.path.abspath(__file__)) | |
| # The two heads of the Laplacian pyramid, and the scale each one produces. | |
| OUTPUTS = [("NCHW_output_2x", 2), ("NCHW_output_4x", 4)] | |
| def upsample(net, img, node_name, scale): | |
| """Run one pyramid head and rebuild a colour image from its luma output. | |
| Mirrors dnn_superres' preprocess_YCrCb / reconstruct_YCrCb: the network only | |
| ever sees the Y channel, and Cr/Cb are bicubically upscaled and merged back. | |
| """ | |
| ycrcb = cv.cvtColor(img, cv.COLOR_BGR2YCrCb).astype(np.float32) / 255.0 | |
| y, cr, cb = cv.split(ycrcb) | |
| # This model takes NHWC [1, H, W, 1] even though its outputs are NCHW. | |
| net.setInput(y.reshape(1, y.shape[0], y.shape[1], 1)) | |
| y_hr = net.forward(node_name)[0, 0] | |
| cr_hr = cv.resize(cr, None, fx=scale, fy=scale) | |
| cb_hr = cv.resize(cb, None, fx=scale, fy=scale) | |
| merged = cv.merge([y_hr, cr_hr, cb_hr]) | |
| # np.rint, not a plain cast: convertTo(CV_8U) rounds, and truncating here | |
| # would drift from the C++ demo by a few levels after YCrCb -> BGR. | |
| merged_8u = np.rint(merged * 255.0).clip(0, 255).astype(np.uint8) | |
| return cv.cvtColor(merged_8u, cv.COLOR_YCrCb2BGR) | |
| def main(): | |
| parser = argparse.ArgumentParser(description="LapSRN x4 super-resolution (ONNX) demo") | |
| parser.add_argument("--model", default=os.path.join(here, "lapsrn_x4_2026sep.onnx")) | |
| parser.add_argument("--image", default=os.path.join(here, "example_outputs", "input_image.png")) | |
| parser.add_argument("--output-dir", default=os.path.join(here, "example_outputs")) | |
| parser.add_argument("--scale", type=int, choices=[2, 4], help="only run this scale (default: both)") | |
| args = parser.parse_args() | |
| img = cv.imread(args.image) | |
| if img is None: | |
| raise SystemExit("could not read image: %s" % args.image) | |
| net = cv.dnn.readNetFromONNX(args.model) | |
| wanted = [(n, s) for n, s in OUTPUTS if args.scale in (None, s)] | |
| print("lapsrn_x4 input %dx%d" % (img.shape[1], img.shape[0])) | |
| for node_name, scale in wanted: | |
| out = upsample(net, img, node_name, scale) | |
| path = os.path.join(args.output_dir, "output_image_%dx.png" % scale) | |
| cv.imwrite(path, out) | |
| print(" %-16s x%d -> %dx%d %s" % (node_name, scale, out.shape[1], out.shape[0], path)) | |
| if __name__ == "__main__": | |
| main() | |