-
Notifications
You must be signed in to change notification settings - Fork 0
/
log
47 lines (36 loc) · 2.06 KB
/
log
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
1. version 0.0
New file structure has been designed for efficient work. Tye. 2020/11/30
2. version 0.1
TestNoseLocation.py: [modified from TestHeadSegmentation.py] Mark, 2020/12/2
util.make_video.py: [create] Mark, 2020/12/2
TestClassification.py: [modify] Mark, 2020/12/2
Note 1: voting under equivalence class TO BE ADDED in TestClassification.py.
Note 2: Dynamic updating mode is applied in both TestNoseLocation.py and TestClassification.py
3. version 0.2
main.py: [created] Tye, 2020/12/03, automatically record results by log file.
dataset.py [modified] Tye, 2020/12/03, merge "get_yaw" into the body of the "PlaneDataset" CLASS.
The project is re-organized and now can conduct classifying and locating automatically.
Some modifications have been made to adapt to current version, especially CLASS in TestClassification.py and SegmentScene.py
4. version 0.3
pointnet.py: [modified] Tye, 2020/12/04,
add "PointNetClsSeg" CLASS,
change the way of extracting point-wise features
dgcnn.py: [modified] Tye, 2020/12/04,
add "DGCNNClsSeg" CLASS,
change the way of extracting global and point-wise features
net.py: [modified] Tye, 2020/12/04,
define the cls-head and seg-head in different backbones,
rewrite the "get_loss" CLASS, where we add the smoothed cross-entropy loss
stn.py: [created] Tye, 2020/12/04,
it contains the subnetworks for transformation. "STN3d", "STNkd" and "TransformNet" are declared.
"TransformNet" is a special version designed for DGCNN.
parts_gt.py: [created] Mark, 2020/12/04,
it receives part labels and return the form suiting CE Loss of point-wise classification.
5 version 0.4
TestPartSegmentation.py [created] Mark, 2020/12/05,
this file is added for eval multi-parts segmentation and its visualization
TestNoseLocation.py [modified] Mark, 2020/12/05,
The original predicted_head is substituted to suit the multi-parts segmentation
Note: the new cls loss raise errors (solved by Ybb)
Note: if we load gt in dataloader, it suffers from .bin loading error
Note: