modInference.py

import numpy as np
import cv2
import torch
import glob as glob
import os
import time

import matplotlib.pyplot as plt

from utils.annotations import CNNpostAnnotations
#from utils.annotations import inference_annotations
from utils.transforms import infer_transforms

def main(CNN, model, input):
    np.random.seed(42)

    image = input
    orig_image = image.copy()
    image = cv2.cvtColor(orig_image, cv2.COLOR_BGR2RGB)
    image = infer_transforms(image)
    image = torch.unsqueeze(image, 0)

    DEVICE = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
    CLASSES = ['__background__', 'Cell']

    outputs = model(image.to(DEVICE))
    
    # Load all detection to CPU for further operations.
    outputs = [{k: v.to('cpu') for k, v in t.items()} for t in outputs]
    print(outputs)
    # Carry further only if there are detected boxes.
    if len(outputs[0]['boxes']) != 0:
        # orig_image = inference_annotations(
        #     outputs, 0.3, CLASSES,
        #     (255, 255, 255), orig_image
        # )
        orig_image, cellImgs = CNNpostAnnotations(
            outputs, 0.3, CLASSES,
            (255, 255, 255), orig_image, CNN
        )
    return orig_image, cellImgs

    cv2.destroyAllWindows()