AI for "object" recognition in low resolution images - PYTHON + KERAS

Hi developers! My company needs a python module that can recognize two classes of "objects" in low-resolution images. The "objects" to be recognized are specific characteristics (anomalies) of the photographed surface. The stack is python + keras. The developer will receive a few thousand images already labeled for the two classes of anomalies and will have to create a neural network (of his choice) and two functions for training and use the NN. If necessary, he will have to process the images before training to improve recognition performance. At the end of the work, two python functions (or two notebooks containing them) must be delivered. The pseudo code of this functions is as follows: # TRAINING FUNCTION def train(images_folder_path): # Load image from "images_folder_path" ... #Prepare images for training (optional, only if is needed) ... # Define NN model parameters ... # Train NN ... # Save NN model and weight # RECOGNITION FUNCTION def recognition(path_of_a_single_image, path_NN_model, path_NN_weights): # Load NN parameter from file "path_NN_model" ... # Load NN weights from file "path_NN_weights" ... # Load image from "path_of_single_image" ... # Search object in the image ... # Return an array with position and class of finded object return [ # Recognized object 1 {x:"", y""", width:"", height:"", class:"0"}, # Recognized object 2 {x:"", y""", width:"", height:"", class:"0"}, # Recognized object 3 {x:"", y""", width:"", height:"", class:"1"}, ]