main.py

import gradio as gr
from PIL import Image

import sys
#Set the Working Cloned directory - To find local version of the library
sys.path.append('/home/user/app/Mask_RCNN') 

import mrcnn.model as modellib
from mrcnn.config import Config
import pickle


def get_predictions_from_mask_rcnn(mask_model, image):

    predictions_mask_rcnn = []

    # Run Object Detection using Mask RCNN
    results = mask_model.detect(image, verbose=0) 
    
    # Get the Results
    r = results[0]
    
    # Get Predictions ROIs, Class IDs, Scores and Masks
    pred_boxes = r['rois']
    pred_class_ids = r['class_ids']
    pred_scores = r['scores']
    pred_masks = r['masks']
    
    print(f"predictions - rois shape: {pred_boxes.shape}, scores: {pred_scores}, masks shape: {pred_masks.shape}")
    
    # Update the Predictions
    predictions_mask_rcnn.append((pred_boxes, pred_class_ids, pred_scores, pred_masks))
                
    # Finally, return the Predictions
    return predictions_mask_rcnn

# Load Config        
def load_config(filepath):
    with open(filepath, 'rb') as f:
        return pickle.load(f)
    
class RSNA_Detect_Config(Config):
    """Configuration for training pneumonia detection on the RSNA pneumonia dataset
    Overrides values in the base Config class
    """
    
    # Provided the configuration a recognizable name  
    NAME = 'RSNA_pneumonia'
    
    # Train on 1 GPU and 8 images per GPU. We can put multiple images on each
    # GPU because the images are small. Batch size is 8 (GPUs * images/GPU).
    GPU_COUNT = 1

    BACKBONE = 'resnet50'
    
    NUM_CLASSES = 2                                 # background + 1 pneumonia classes
    
    IMAGE_MIN_DIM = 256                             # default  800
    IMAGE_MAX_DIM = 256                             # default  1024
    RPN_ANCHOR_SCALES = (16, 32, 64, 128, 256)      # default  (32, 64, 128, 256, 512)
    TRAIN_ROIS_PER_IMAGE = 32                       # default  200
    MAX_GT_INSTANCES = 4                            # default  100
    DETECTION_MAX_INSTANCES = 4                     # default  100, to match distribution
    DETECTION_MIN_CONFIDENCE = 0.78                 # match the target distribution
    DETECTION_NMS_THRESHOLD = 0.01                  # default 0.3

# Get object reference for RSNA_Eval_Config
config = RSNA_Detect_Config()

mask_rcnn_inference_model = modellib.MaskRCNN(mode = 'inference', 
                                              config =  config, 
                                              model_dir = "assests")

def predict(inp):
    confidences = {"cat": 0.0, "dog": 0.5}
    predictions = get_predictions_from_mask_rcnn(mask_rcnn_inference_model, inp)
    print(predictions)
    return inp, confidences

gr.Interface(fn=predict, 
             inputs=gr.Image(type="pil"),
             outputs=[gr.Image(type="pil"), gr.Label(num_top_classes=3)],
             examples=[]).launch()