Question mAP and FPS

[wenjun90]

Dir sxhxliang,

Could I ask you the mAP and FPS you obtain with detectron_efficientNet on COCO dataset?

Thank you

[sxhxliang]

build_fcos_efficientnet_fpn_backbone
epoch:6 batchsize:8 gpus:4 coco results:

AP	AP50	AP75	APs	APm	APl
29.017	45.281	30.911	16.333	31.203	38.439

[wenjun90]

Could you share me FPS ou prediction time, please? Thank you so much.

[sxhxliang]

sorry, I have no time to do this, it is easy to measure the speed and you can test the model yourself.

def compute_speed(model, input_size, device, iteration):
    print('input_size', input_size)
    torch.cuda.set_device(device)
    torch.backends.cudnn.benchmark = True

    model.eval()
    model = model.cuda()

    input = torch.randn(*input_size).cuda()

    for _ in range(10):
        model(input)

    print('=========Speed Testing=========')
    torch.cuda.synchronize()
    torch.cuda.synchronize()
    t_start = time.time()
    for _ in range(iteration):
        model(input)
    torch.cuda.synchronize()
    torch.cuda.synchronize()
    elapsed_time = time.time() - t_start
    print(
        'Elapsed time: [%.2f s / %d iter]' % (elapsed_time, iteration))
    print('Speed Time: %.2f ms / iter    FPS: %.2f' % (
        elapsed_time / iteration * 1000, iteration / elapsed_time))