submit

GFPGAN.py

@@ -0,0 +1,38 @@
+import cv2
+# import torch
+import onnxruntime
+import numpy as np
+
+# gfpgan converted to onnx
+# using https://github.com/xuanandsix/GFPGAN-onnxruntime-demo
+# same inference code for GFPGANv1.2, GFPGANv1.3, GFPGANv1.4
+
+class GFPGAN:
+    def __init__(self, model_path="GFPGANv1.4.onnx", device='cpu'):
+        session_options = onnxruntime.SessionOptions()
+        session_options.graph_optimization_level = onnxruntime.GraphOptimizationLevel.ORT_ENABLE_ALL
+        providers = ["CPUExecutionProvider"]
+        if device == 'cuda':
+            providers = [("CUDAExecutionProvider", {"cudnn_conv_algo_search": "DEFAULT"}),"CPUExecutionProvider"]
+        self.session = onnxruntime.InferenceSession(model_path, sess_options=session_options, providers=providers)
+        self.resolution = self.session.get_inputs()[0].shape[-2:]
+
+    def preprocess(self, img):
+        img = cv2.resize(img, self.resolution, interpolation=cv2.INTER_LINEAR)
+        img = img.astype(np.float32)[:,:,::-1] / 255.0
+        img = img.transpose((2, 0, 1))
+        img = (img - 0.5) / 0.5
+        img = np.expand_dims(img, axis=0).astype(np.float32)
+        return img
+
+    def postprocess(self, img):
+        img = (img.transpose(1,2,0).clip(-1,1) + 1) * 0.5
+        img = (img * 255)[:,:,::-1]
+        img = img.clip(0, 255).astype('uint8')
+        return img
+
+    def enhance(self, img):
+        img = self.preprocess(img)
+        output = self.session.run(None, {'input':img})[0][0]
+        output = self.postprocess(output)
+        return output

GPEN.py

@@ -0,0 +1,34 @@
+import cv2
+# import torch
+import onnxruntime
+import numpy as np
+
+class GPEN:
+    def __init__(self, model_path="GPEN-BFR-512.onnx", device='cpu'):
+        session_options = onnxruntime.SessionOptions()
+        session_options.graph_optimization_level = onnxruntime.GraphOptimizationLevel.ORT_ENABLE_ALL
+        providers = ["CPUExecutionProvider"]
+        if device == 'cuda':
+            providers = [("CUDAExecutionProvider", {"cudnn_conv_algo_search": "DEFAULT"}),"CPUExecutionProvider"]
+        self.session = onnxruntime.InferenceSession(model_path, sess_options=session_options, providers=providers)
+        self.resolution = self.session.get_inputs()[0].shape[-2:]
+
+    def preprocess(self, img):
+        img = cv2.resize(img, self.resolution, interpolation=cv2.INTER_LINEAR)
+        img = img.astype(np.float32)[:,:,::-1] / 255.0
+        img = img.transpose((2, 0, 1))
+        img = (img - 0.5) / 0.5
+        img = np.expand_dims(img, axis=0).astype(np.float32)
+        return img
+
+    def postprocess(self, img):
+        img = (img.transpose(1,2,0).clip(-1,1) + 1) * 0.5
+        img = (img * 255)[:,:,::-1]
+        img = img.clip(0, 255).astype('uint8')
+        return img
+
+    def enhance(self, img):
+        img = self.preprocess(img)
+        output = self.session.run(None, {'input':img})[0][0]
+        output = self.postprocess(output)
+        return output

LICENSE

@@ -0,0 +1,22 @@
+MIT License
+Copyright (c) [2023] [Harisreedhar]
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+**The Software may include models or any other intellectual property that are owned and licensed separately by their respective authors or copyright holders. The ownership and licensing terms of these components supersede the terms of this MIT License. It is your responsibility to review and comply with the individual licensing terms of such components.**
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

README.md

@@ -1,2 +1,6 @@
 # Face-Upscalers-ONNX
 ONNX-Powered Inference for State-of-the-Art Face Upscalers
+
+## MIT License for Repository Codes
+
+The codebase in this repository, **excluding the converted models**, is licensed under the permissive MIT License. This means you can freely use, modify, and distribute the code in this repository without any restrictions, provided that you include the original copyright notice and disclaimer.

codeformer.py

@@ -0,0 +1,38 @@
+import cv2
+# import torch
+import onnxruntime
+import numpy as np
+
+# codeformer converted to onnx
+# using https://github.com/redthing1/CodeFormer
+
+class CodeFormer:
+    def __init__(self, model_path="codeformer.onnx", device='cpu'):
+        session_options = onnxruntime.SessionOptions()
+        session_options.graph_optimization_level = onnxruntime.GraphOptimizationLevel.ORT_ENABLE_ALL
+        providers = ["CPUExecutionProvider"]
+        if device == 'cuda':
+            providers = [("CUDAExecutionProvider", {"cudnn_conv_algo_search": "DEFAULT"}),"CPUExecutionProvider"]
+        self.session = onnxruntime.InferenceSession(model_path, sess_options=session_options, providers=providers)
+        self.resolution = self.session.get_inputs()[0].shape[-2:]
+
+    def preprocess(self, img, w):
+        img = cv2.resize(img, self.resolution, interpolation=cv2.INTER_LINEAR)
+        img = img.astype(np.float32)[:,:,::-1] / 255.0
+        img = img.transpose((2, 0, 1))
+        img = (img - 0.5) / 0.5
+        img = np.expand_dims(img, axis=0).astype(np.float32)
+        w = np.array([w], dtype=np.double)
+        return img, w
+
+    def postprocess(self, img):
+        img = (img.transpose(1,2,0).clip(-1,1) + 1) * 0.5
+        img = (img * 255)[:,:,::-1]
+        img = img.clip(0, 255).astype('uint8')
+        return img
+
+    def enhance(self, img, w=0.9):
+        img, w = self.preprocess(img, w)
+        output = self.session.run(None, {'x':img, 'w':w})[0][0]
+        output = self.postprocess(output)
+        return output