【语义分割】——FCN测试

项目: 采用mmsegmentation

下载预训练的模型： from fcn

硬件条件

• 1070单卡（8GB）
• cuda10.1，cudnn7.6.3

1. 测试

配置好相关的环境后，采用官方的测试demo脚本。get_start

from mmseg.apis import inference_segmentor, init_segmentor
import mmcv

config_file = 'configs/pspnet/pspnet_r50-d8_512x1024_40k_cityscapes.py'
checkpoint_file = 'checkpoints/pspnet_r50-d8_512x1024_40k_cityscapes_20200605_003338-2966598c.pth'

# build the model from a config file and a checkpoint file
model = init_segmentor(config_file, checkpoint_file, device='cuda:0')

# test a single image and show the results
img = 'test.jpg'  # or img = mmcv.imread(img), which will only load it once
result = inference_segmentor(model, img)
# visualize the results in a new window
model.show_result(img, result, show=True)
# or save the visualization results to image files
model.show_result(img, result, out_file='result.jpg')

# test a video and show the results
video = mmcv.VideoReader('video.mp4')
for frame in video:
    result = inference_segmentor(model, frame)
    model.show_result(frame, result, wait_time=1)

• 视频测试源:MOT-16

2. 速度/可视化结果

Cityscapes

Method	Backbone	total_iters	Crop Size	Inf time (fps)
FCN	R-50-D8	40k	512×1024	1.431

ADE20K

Method	Backbone	total_iters	Crop Size	Inf time (fps)
FCN	R-50-D8	80k	512×512	4.371

Pascal VOC 2012 + Aug

Method	Backbone	total_iters	Crop Size	Inf time (fps)
FCN	R-50-D8	20k	512×512	5.214

总结

• 同样的模型，对比来说，使用最后一个数据集训练的效果是最好的。分割效果+干扰。

other

• 原理+训练