GPU fps memory

最新推荐文章于 2022-12-27 17:51:22 发布

LLdty

最新推荐文章于 2022-12-27 17:51:22 发布

阅读量245

点赞数

本文链接：https://blog.csdn.net/weixin_37859224/article/details/117328450

版权

python

def predict_image(detector):
    if FLAGS.run_benchmark:
        detector.predict(
            FLAGS.image_file,
            FLAGS.threshold,
            warmup=100,
            repeats=100,
            run_benchmark=True)
    else:
        imgs_lists =  get_image_list(FLAGS.image_file)
        os.system('nvidia-smi -q -d Memory |grep -A4 GPU|grep Free >tmp')
        memory_gpu = [int(x.split()[2]) for x in open('tmp', 'r').readlines()]
        os.system('rm tmp')
        GPU_free1 = memory_gpu[FLAGS.GPU_id]
        # print('GPU:%d - free: %s' % (FLAGS.GPU_id, str(GPU_free1)))
        GPU_free2 = 0
        result_save_path = os.path.join(FLAGS.output_dir, 'result.txt')
        with open(result_save_path,'w') as f:
            time_start = time.time()
            for i,img in enumerate(imgs_lists):
                results = detector.predict(img, FLAGS.threshold)
                # os.system('nvidia-smi -q -d Memory |grep -A4 GPU|grep Free >tmp')
                # memory_gpu = [int(x.split()[2]) for x in open('tmp', 'r').readlines()]
                # os.system('rm tmp')
                # GPU_free = memory_gpu[FLAGS.GPU_id]
                # print('GPU:%d - free: %s' % (FLAGS.GPU_id, str(GPU_free)))
                if i == len(imgs_lists)-1:
                    os.system('nvidia-smi -q -d Memory |grep -A4 GPU|grep Free >tmp')
                    memory_gpu = [int(x.split()[2]) for x in open('tmp', 'r').readlines()]
                    os.system('rm tmp')
                    GPU_free2 = memory_gpu[FLAGS.GPU_id]
                dict = {}
                dict['image_name']=img.split('.jpg')[0].split('/')[-1]+'.jpg'
                # image = cv2.imread(img)
                # dict['width']=image.shape[1]
                # dict['height']=image.shape[0]
                # dict['bbox']=[]
                classes = []
                bboxes = []
                for dt in results['boxes']:
                    clsid, bbox, score = int(dt[0]), list(dt[2:]), float(dt[1])
                    classes.append(detector.config.labels[clsid])
                    bboxes.append(bbox)
                    # obj={}
                    # obj['label']=detector.config.labels[clsid]
                    # obj['bbox']=bbox
                    # obj['rate']=score
                    # dict['bbox'].append(obj)
                dict['classes'] = classes
                dict['bboxes'] = bboxes
                f.write(json.dumps(dict,cls=NpEncoder)+'\n')
                visualize(
                    img,
                    results,
                    detector.config.labels,
                    mask_resolution=detector.config.mask_resolution,
                    output_dir=FLAGS.output_dir,
                    threshold=FLAGS.threshold)
            time_fn = time.time()
            print('-----------------------------------------------------------------')
            print('GPU:%d - Consumption: %s M' % (FLAGS.GPU_id, int(GPU_free1)-int(GPU_free2)))
            ms = (time_fn - time_start) * 1000.0 / len(imgs_lists)
            print("Inference: {} ms per batch image".format(ms))