Spaces:
Build error
Build error
File size: 2,407 Bytes
d7a991a |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 |
#!/usr/bin/env bash
# Copyright (c) OpenMMLab. All rights reserved.
import argparse
import time
import torch
from mmcv import Config
from mmcv.cnn import fuse_conv_bn
from mmcv.parallel import MMDataParallel
from mmcv.runner.fp16_utils import wrap_fp16_model
from mmpose.datasets import build_dataloader, build_dataset
from mmpose.models import build_posenet
def parse_args():
parser = argparse.ArgumentParser(
description='MMPose benchmark a recognizer')
parser.add_argument('config', help='test config file path')
parser.add_argument('--bz', default=32, type=int, help='test config file path')
args = parser.parse_args()
return args
def main():
args = parse_args()
cfg = Config.fromfile(args.config)
# Since we only care about the forward speed of the network
cfg.model.pretrained=None
cfg.model.test_cfg.flip_test=False
cfg.model.test_cfg.use_udp=False
cfg.model.test_cfg.post_process='none'
# set cudnn_benchmark
if cfg.get('cudnn_benchmark', False):
torch.backends.cudnn.benchmark = True
# build the dataloader
dataset = build_dataset(cfg.data.val)
data_loader = build_dataloader(
dataset,
samples_per_gpu=args.bz,
workers_per_gpu=cfg.data.workers_per_gpu,
dist=False,
shuffle=False)
# build the model and load checkpoint
model = build_posenet(cfg.model)
model = MMDataParallel(model, device_ids=[0])
model.eval()
# get the example data
for i, data in enumerate(data_loader):
break
# the first several iterations may be very slow so skip them
num_warmup = 100
inference_times = 100
with torch.no_grad():
start_time = time.perf_counter()
for i in range(num_warmup):
torch.cuda.synchronize()
model(return_loss=False, **data)
torch.cuda.synchronize()
elapsed = time.perf_counter() - start_time
print(f'warmup cost {elapsed} time')
start_time = time.perf_counter()
for i in range(inference_times):
torch.cuda.synchronize()
model(return_loss=False, **data)
torch.cuda.synchronize()
elapsed = time.perf_counter() - start_time
fps = args.bz * inference_times / elapsed
print(f'the fps is {fps}')
if __name__ == '__main__':
main()
|