Bad Result at demo_folder.py
Closed this issue · 5 comments
I tested based on https://github.com/L-Reichardt/Cylinder3D_spconv_v2 repo. Cause I use CUDA 11.3 and Spconv1.x is deprecated.
I run demo_folder.py and I got below result which is embarrassing.
matched parameter sets: 226, and no matched: 48
Validation per class iou:
car : 0.00%
bicycle : 0.00%
motorcycle : 0.00%
truck : 0.00%
bus : 0.00%
person : 0.00%
bicyclist : 0.00%
motorcyclist : 0.00%
road : 0.00%
parking : 0.00%
sidewalk : 0.00%
other-ground : 0.00%
building : 0.00%
fence : 0.00%
vegetation : 30.61%
trunk : 0.00%
terrain : 0.00%
pole : 0.00%
traffic-sign : 0.00%
Current val miou is 1.611
Current val loss is 3.492
What did I do wrong??
same problem,how to solve it?
@cama36 https://github.com/L-Reichardt/Cylinder3D_spconv_v2. In here author upload new WEIGHT. You can refer to it.
and you can expect below result which is slight lower performance than original paper.
Validation per class iou:
car : 96.92%
bicycle : 54.25%
motorcycle : 80.25%
truck : 33.20%
bus : 69.36%
person : 45.18%
bicyclist : 59.40%
motorcyclist : 51.25%
road : 95.75%
parking : 55.55%
sidewalk : 86.47%
other-ground : 0.00%
building : 80.21%
fence : 37.01%
vegetation : 89.62%
trunk : 73.00%
terrain : 80.73%
pole : 73.18%
traffic-sign : 73.67%
Current val miou is 65.000
Current val loss is 0.600
@DonghunP
The files in the google drive, though they are .pt file, when downloaded they are a zip folder, how did you get around this ?
Eager to hear from you soon :)
hi @DonghoonPark12
could you please share the weights for the https://github.com/L-Reichardt/Cylinder3D-updated-CUDA repo? The link seems to be broken.
Thanks
I met with similar problem
(com88) adminpc@adminpc-B760-DS3H:/media/adminpc/828F225A817AD4FA/Cylinder3D$ python demo_folder.py --demo-folder ./dataset/sequences/00/velodyne/ --save-folder ./model_save_dir --demo-label-folder ./dataset/sequences/00/labels/
......
save ./model_save_dir/000000.label
save ./model_save_dir/000001.label
save ./model_save_dir/000002.label
......
save ./model_save_dir/004539.label
save ./model_save_dir/004540.label
Validation per class iou:
car : 0.00%
bicycle : 0.00%
motorcycle : 0.00%
truck : 0.00%
bus : 0.24%
person : 0.00%
bicyclist : 0.00%
motorcyclist : 0.00%
road : 0.00%
parking : 0.00%
sidewalk : 0.00%
other-ground : 0.00%
building : 0.00%
fence : 0.00%
vegetation : 0.00%
trunk : 0.00%
terrain : 0.00%
pole : 0.00%
traffic-sign : 0.00%
Current val miou is 0.013
Current val loss is 3.951