import gradio as gr import numpy as np import json import os from PIL import Image import onnxruntime as rt class ONNXModel: def __init__(self, dir_path) -> None: """Method to get name of model file. Assumes model is in the parent directory for script.""" model_dir = os.path.dirname(dir_path) with open(os.path.join(model_dir, "signature.json"), "r") as f: self.signature = json.load(f) self.model_file = os.path.join(model_dir, self.signature.get("filename")) if not os.path.isfile(self.model_file): raise FileNotFoundError(f"Model file does not exist") # get the signature for model inputs and outputs self.signature_inputs = self.signature.get("inputs") self.signature_outputs = self.signature.get("outputs") self.session = None if "Image" not in self.signature_inputs: raise ValueError("ONNX model doesn't have 'Image' input! Check signature.json, and please report issue to Lobe.") # Look for the version in signature file. # If it's not found or the doesn't match expected, print a message version = self.signature.get("export_model_version") if version is None or version != EXPORT_MODEL_VERSION: print( f"There has been a change to the model format. Please use a model with a signature 'export_model_version' that matches {EXPORT_MODEL_VERSION}." ) def load(self) -> None: """Load the model from path to model file""" # Load ONNX model as session. self.session = rt.InferenceSession(path_or_bytes=self.model_file) def predict(self, image: Image.Image) -> dict: """ Predict with the ONNX session! """ # process image to be compatible with the model img = self.process_image(image, self.signature_inputs.get("Image").get("shape")) # run the model! fetches = [(key, value.get("name")) for key, value in self.signature_outputs.items()] # make the image a batch of 1 feed = {self.signature_inputs.get("Image").get("name"): [img]} outputs = self.session.run(output_names=[name for (_, name) in fetches], input_feed=feed) return self.process_output(fetches, outputs) def process_image(self, image: Image.Image, input_shape: list) -> np.ndarray: """ Given a PIL Image, center square crop and resize to fit the expected model input, and convert from [0,255] to [0,1] values. """ width, height = image.size # ensure image type is compatible with model and convert if not if image.mode != "RGB": image = image.convert("RGB") # center crop image (you can substitute any other method to make a square image, such as just resizing or padding edges with 0) if width != height: square_size = min(width, height) left = (width - square_size) / 2 top = (height - square_size) / 2 right = (width + square_size) / 2 bottom = (height + square_size) / 2 # Crop the center of the image image = image.crop((left, top, right, bottom)) # now the image is square, resize it to be the right shape for the model input input_width, input_height = input_shape[1:3] if image.width != input_width or image.height != input_height: image = image.resize((input_width, input_height)) # make 0-1 float instead of 0-255 int (that PIL Image loads by default) image = np.asarray(image) / 255.0 # format input as model expects return image.astype(np.float32) def process_output(self, fetches: dict, outputs: dict) -> dict: # un-batch since we ran an image with batch size of 1, # convert to normal python types with tolist(), and convert any byte strings to normal strings with .decode() out_keys = ["label", "confidence"] results = {} for i, (key, _) in enumerate(fetches): val = outputs[i].tolist()[0] if isinstance(val, bytes): val = val.decode() results[key] = val confs = results["Confidences"] labels = self.signature.get("classes").get("Label") output = [dict(zip(out_keys, group)) for group in zip(labels, confs)] sorted_output = {"predictions": sorted(output, key=lambda k: k["confidence"], reverse=True)} return sorted_output EXPORT_MODEL_VERSION=1 model = ONNXModel(dir_path="../model.onnx") model.load() def predict(image): image = Image.fromarray(np.uint8(image), 'RGB') prediction = model.predict(image) for output in prediction["predictions"]: output["confidence"] = round(output["confidence"], 2) return prediction inputs = gr.inputs.Image(type="pil") outputs = gr.outputs.JSON() gr.Interface(fn=predict, inputs=inputs, outputs=outputs).launch()