文件名称：ICNet-pytorch:由pytorch实现的ICNet，用于在高分辨率图像上进行实时语义分割，在城市景观上，mIOU = 71.0，单次推理时间为19ms，FPS为52.6
文件大小：20.54MB
文件格式：ZIP
更新时间：2024-05-23 08:34:21
real-time pytorch semantic-segmentation cityscapes icnet
描述 此回购包含ICNet实现由PyTorch，基于的Hengshuang赵和等。 al（ECCV'18）。 默认情况下，对进行培训和评估。 要求 带有以下pip3 install -r requirements.txt Python 3.6或更高版本： 火炬== 1.1.0 torchsummary == 1.5.1 火炬视觉== 0.3.0 numpy == 1.17.0 枕头== 6.0.0 PyYAML == 5.1.2 更新 2019.11.15：更改crop_size=960 ，最佳mIoU增至71.0％。 花了大约2天的时间。 获取 表现 方法 浓度（％） 时间（毫秒） 第一人称射击 内存（GB） 显卡 ICNet（论文） 67.7％ 33毫秒 30.3 1.6 泰坦X ICNet（我们的） 71.0％ 19毫秒 52.6 1.86 GTX
【文件预览】：
ICNet-pytorch-master
----.gitignore(1KB)
----requirements.txt(138B)
----ckpt()
--------.gitignore(16B)
--------icnet_resnet50_log.txt(1.63MB)
--------icnet_resnet50_evaluate_log.txt(58KB)
----dataset()
--------segbase.py(4KB)
--------.gitignore(22B)
--------cityscapes.py(7KB)
--------__init__.py(88B)
----models()
--------segbase.py(2KB)
--------__init__.py(24B)
--------icnet.py(5KB)
--------model_store.py(2KB)
--------base_models()
----LICENSE(1KB)
----utils()
--------__init__.py(206B)
--------download.py(3KB)
--------visualize.py(6KB)
--------lr_scheduler.py(778B)
--------loss.py(1KB)
--------metric.py(6KB)
--------logger.py(1KB)
----README.md(6KB)
----configs()
--------.gitignore(20B)
--------icnet.yaml(872B)
----demo()
--------munster_000150_000019_leftImg8bit_mIoU_0.695.png(17KB)
--------munster_000150_000019_leftImg8bit_label.png(21KB)
--------munster_000158_000019_leftImg8bit_label.png(20KB)
--------munster_000061_000019_leftImg8bit_label.png(27KB)
--------munster_000061_000019_leftImg8bit_src.png(2.14MB)
--------munster_000121_000019_leftImg8bit_label.png(18KB)
--------frankfurt_000001_057181_leftImg8bit_src.png(2.12MB)
--------munster_000075_000019_leftImg8bit_mIoU_0.690.png(12KB)
--------frankfurt_000001_057181_leftImg8bit_mIoU_0.727.png(24KB)
--------munster_000061_000019_leftImg8bit_mIoU_0.672.png(25KB)
--------munster_000106_000019_leftImg8bit_mIoU_0.672.png(19KB)
--------munster_000158_000019_leftImg8bit_mIoU_0.692.png(18KB)
--------munster_000061_000019_leftImg8bit_mIoU_0.704.png(23KB)
--------munster_000121_000019_leftImg8bit_mIoU_0.694.png(15KB)
--------munster_000124_000019_leftImg8bit_src.png(2.11MB)
--------munster_000124_000019_leftImg8bit_mIoU_0.695.png(17KB)
--------munster_000121_000019_leftImg8bit_mIoU_0.678.png(14KB)
--------frankfurt_000001_057181_leftImg8bit_mIoU_0.680.png(27KB)
--------munster_000124_000019_leftImg8bit_mIoU_0.660.png(18KB)
--------munster_000121_000019_leftImg8bit_src.png(2.07MB)
--------munster_000158_000019_leftImg8bit_mIoU_0.658.png(18KB)
--------munster_000121_000019_leftImg8bit_mIoU_0.660.png(16KB)
--------lindau_000005_000019_leftImg8bit_mIoU_0.657.png(19KB)
--------munster_000158_000019_leftImg8bit_mIoU_0.676.png(17KB)
--------munster_000124_000019_leftImg8bit_mIoU_0.696.png(17KB)
--------munster_000075_000019_leftImg8bit_src.png(2.04MB)
--------munster_000124_000019_leftImg8bit_label.png(19KB)
--------frankfurt_000001_057181_leftImg8bit_mIoU_0.716.png(24KB)
--------munster_000150_000019_leftImg8bit_src.png(2.34MB)
--------lindau_000005_000019_leftImg8bit_src.png(2.13MB)
--------lindau_000005_000019_leftImg8bit_label.png(22KB)
--------munster_000061_000019_leftImg8bit_mIoU_0.692.png(24KB)
--------munster_000150_000019_leftImg8bit_mIoU_0.696.png(18KB)
--------munster_000106_000019_leftImg8bit_mIoU_0.690.png(18KB)
--------munster_000106_000019_leftImg8bit_src.png(2.18MB)
--------munster_000075_000019_leftImg8bit_mIoU_0.703.png(12KB)
--------munster_000158_000019_leftImg8bit_src.png(2.1MB)
--------lindau_000005_000019_leftImg8bit_mIoU_0.700.png(19KB)
--------lindau_000005_000019_leftImg8bit_mIoU_0.705.png(19KB)
--------munster_000075_000019_leftImg8bit_label.png(17KB)
--------munster_000075_000019_leftImg8bit_mIoU_0.672.png(13KB)
--------munster_000150_000019_leftImg8bit_mIoU_0.660.png(18KB)
--------munster_000106_000019_leftImg8bit_label.png(20KB)
--------frankfurt_000001_057181_leftImg8bit_label.png(29KB)
--------munster_000106_000019_leftImg8bit_mIoU_0.703.png(18KB)
----ICNet.png(446KB)
----evaluate.py(8KB)
----train.py(10KB)