-
Notifications
You must be signed in to change notification settings - Fork 0
/
Update.log
119 lines (117 loc) · 3.67 KB
/
Update.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
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
* 2019.05.19
- Add TODO: Run-length encoding for masks
* 2019.05.06
- This project is gonna pause for a while (maybe 1 or 2 month)
- Waiting for DeepFashion2
- Learning Mask_RCNN
- Other project/repo needs me
* 2019.05.05
- Start learning Mask_RCNN code
- Waiting for DeepFashion2
* 2019.05.01
- Use `cdict` instead of `dot` to measure the distance between vgg/resnet/densenet vector
- A litter better result, but still not good
* 2019.04.30
- Try to distinguish person w/o skis/snowboard by distance
- Failed because the distance distribution is not bimodal
* 2019.04.29
- Set a new goal
* 2019.04.27
- Extract feature from bulk boxes
- Using my image as input vector
- Still not work well
* 2019.04.26
- Update Tools `Show_Img`
- Test Query on random choice
> Not working when use vgg/resnet/densenet vector (w/o L1 norm) to query.
> It seems body-shape/pose take dominant contribution than clothes/color.
> It works better in small datasets, maybe because it can't find similar body-shape/pose with limited images.
---
* 2019.04.24
- Extract feature from bulk masks
* 2019.04.23
- Separate bulk run files
- Datetime analysis
* 2019.04.22
- Wanlong images processing (Finish)
- Merge Mask R-CNN results
* 2019.04.20
- Wanlong images processing (Start)
* 2019.04.19
- Keep working on GPU
- Switch to bulk pipeline
- Add query pipeline
- Start building tools
* 2019.04.17
- Keep working on GPU
- Add Tips for GPU
- BATCH_SIZE -> IMAGES_PER_GPU
- Add VGG and DenseNet
* 2019.04.16
- Start downloading bulk images from @MingxuanHu
- Start coding on GPU node
- Tips for GPU
- Make sure tensorflow-gpu was installed
- Watching GPU memory
* 2019.04.15
- Add Tags and emoji
* 2019.04.14
- Cleanup DR and Clustering Code & Notebook
- Upload Code
* 2019.04.13
- Visualize tSNE clusters
- #TAG-ignore
- only got 8 clusters
- two of them are belong to one cluster
- one of them are mixed with two but similar skiers
- one of them is not clustered by skiers clothes
- Rank person
- #TAG-ignore
- rank_ID = np.argsort(np.dot(inputV, dataV.T))[::-1]
- Acceptable result
- Reference: https://github.com/willard-yuan/flask-keras-cnn-image-retrieval.git
* 2019.04.12
- Use ResNet50 extract feature of InBox Pixels
- Use PCA, tSNE, UMAP to DR and visualize results
- Visualize UMAP cluster InBox Pixels and row images
- Add Images for BoxSize Distribution and Dimensionality Reduction
* 2019.04.11
- Add v0.0 tag message
- Rephrasing, Add some explanations, Add emoji
- Invite Collaborators
* 2019.04.10
- Change cutoff
- Add scores violinplot
- Add InBox Pixels Heatmap
- Upload analysis code
* 2019.04.09
- Update image for result analysis
- Messed up masks shape ...
- Re-run Mask_RCNN
* 2019.04.08
- Save Mask_RCNN outputs for image datasets (`pw_20190331_wl_n1`)
- Pandas v0.23 -> v0.24
- @MingxuanHu sent me his scraping results (more than 6G...)
- Update imgs
- Analysis Mask_RCNN outputs
- Add and Update TODO, Add TAG-HEAD
* 2019.04.04
- Setup Mask_RCNN
- Test Mask_RCNN on demo and real data
- Get URLs for images
- Start Download images
* 2019.04.03
- Test API to get image
- Meet with @MingxuanHu, tell him about API
* 2019.04.02
- Add images
- Rephrasing
- Creat .gitignore
- Creat log
- Add link
- Get some imgs into `SnapData` (ignore)
- Look at Mask R-CNN, try to deploy it
* 2019.04.01
- Creat repo
- Creat README.md
- Commit `intro.`, `Outline`