深度估計方法monodepth2 PSMnet

2023-06-24 06:02:27

monodepth2 預訓練批量測試

雙目測距系列（三）monodepth2模型的測試 https://blog.csdn.net/avideointerfaces/article/details/105647031

官方儲存的是僞彩色圖，儲存為灰階圖：

colormapped_im = (mapper.to_rgba(disp_resized_np)[:, :, 3] * 255).astype(np.uint8)
改
colormapped_im = (mapper.to_rgba(disp_resized_np)[:, :, 0] * 255).astype(np.uint8)

批量處理

python3 slz_test_batch.py --image_path /kitti_07/image_0 --model_name stereo_1024x320

深度估計方法monodepth2 PSMnet

效果

PSMnet 預訓練批量

論文閱讀筆記《Pyramid Stereo Matching Network》https://blog.csdn.net/qq_36104364/article/details/79940969

運作PSMNet網絡時遇到的問題及解決方案

https://blog.csdn.net/qq_36104364/article/details/80406327?utm_medium=distribute.pc_aggpage_search_result.none-task-blog-2~all~first_rank_v2~rank_v25-4-80406327.nonecase&utm_term=psmnet%E9%A2%84%E8%AE%AD%E7%BB%83%E9%AA%8C%E8%AF%81%E5%87%BA%E9%97%AE%E9%A2%98&spm=1000.2123.3001.4430

python3.6 torch1.4 torchvision0.5.0 運作記錄

python Test_img.py --loadmodel weight/pretrained_model_KITTI2015.tar --leftimg ~/dataset/kitti_07/image_0/000000.png --rightimg ~/dataset/kitti_07/image_1/000000.png

報錯;

File "Test_img.py", line 12, in <module>
    from models import *
  File "/home/shilinzhe/PSMNet/models/__init__.py", line 1, in <module>
    from .basic import PSMNet as basic
  File "/home/shilinzhe/PSMNet/models/basic.py", line 8, in <module>
    from submodule import *
ModuleNotFoundError: No module named 'submodule'

猜測是官方要求的python3.7 特性，

改為 from .submodule import * (加個點)

File "/home/shilinzhe/PSMNet/models/stackhourglass.py", line 138
    cost1 = F.upsample(cost1, [self.maxdisp,left.size()[2],left.size()[3]], mode='trilinear')
                                                                                            ^
TabError: inconsistent use of tabs and spaces in indentation

改：把對應位置的tab換空格.....

kitti07 批量處理

python slz_Test_img_batch.py --loadmodel weight/pretrained_model_KITTI2012.tar --leftimg ~/dataset/kitti_07/image_0/ --rightimg ~/dataset/kitti_07/image_1/ --output ~/dataset/kitti_07/disp_psmnet/