Handling requests for a detection problem using Inception V2

Part 1 (2020)
1

For a single image at a time, the detection algorithm works like a charm. But, when I pass multiple images at once, it just crashes.

I had it working on a CPU server earlier by wrapping it around multiprocessing and closing it every time a detection completed.
Here is the code for that:

def process_detection_pipeline(self, front, inputdata):
        try:
            manager = multiprocessing.Manager()
            return_dict = manager.dict()

            def worker_process(return_dic, front, inputdata):
                userid = inputdata['userid']
                if front is not None:
                    frontcrpimg, statusmin, statusmax = self.obj_detectperson.getUserCropImg(front)
                    cv2.imwrite('output/' + str(userid) + '/cropimgs/front.jpg', frontcrpimg)
                    return_dic['1'] = statusmin
                    return_dic['2'] = statusmax
                    return_dic['3'] = frontcrpimg

            p = Process(target=worker_process, args=(return_dict, front, inputdata))
            p.start()
            p.join()
            return return_dict.values()
        except:
            return False, False, None

The same code block is not working in the case of a GPU-based server.
I would like to know if there is a better solution to solve this problem?