/
notes.txt
143 lines (80 loc) · 5.53 KB
/
notes.txt
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
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
render later after fonts are downloaded
active trashcan icon for labels is brighter than images
small new bug, use arrow down/up when labelling first object
when opening up the labeller modal, there is a scrollbar
the label trashcan has a stronger tint than the image
zoom fix - check how it looks after making the original image full height
add gifs
remove unused asset files
social media preview image for https://myvision.ai
check whether the hyperlink to index.html in the readme is enough
when user holds ctrl and scrolls, should zoom-in on the image (be ware of use cases where the user may be drawing or editing shapes)
the remove images modal should be smaller
potentially emit the use of the publicDev folder
potentially decrease the height of the upload datasets modal
upon viewing an image with ml shapes on it and then moving to another one, the shape's fill should disappear
think about caching some of the 'do not show again' modal data, settings
default canvas does appear to have highlight (multiple shape) capabilities, investigate
every 10 mins after the last save, should get a brief notification to remember to save the work
can't use enter in machine learning modal
the labeller overflow still does happen, check if some properties have changed
---------------------------------------------
Things to check on chrome browsers on other pcs:
weird auto horizontal rotate for some images on firefox
potential fix:
img {
image-orientation: from-image;
}
shape labeller modal horizontal and vertical overflow fake borders
label list dropdown horizontal and vertical overflow fake borders
machine learning generated labels horizontal and vertical overflow fake borders
make sure the dim effect is not choppy for slower pcs
-------------------------------------------------
UX:
concider the fact that after editing a label in continuous drawing mode, we should maybe go back to the drawing mode
check whether it is more approriate to add bounding boxes by holding key or clicking it multiple times
highlight shapes to indicate that the user needs to click on them to start adding/removing points - check if really needed
Check to see if the upload images and remove image button are not too high up, causing the user to click remove image when
their actual intention is to remove a shape. To fix this, can potentially move the buttons lower by increasing their padding top
or highlight the image to be removed (in red) on remove image button hover (border or potentially overlay)
Check to see if the new border functionality is not causing users to think that on hovering the delete shape button, the shape would
be highlighted
check if export datasets is enough of an indicator that it is a download functionality
(reason not using download is because the word is long for the popup)
Or maybe having the 'Download Datasets' popover and Export button is clear enough
Change the HTML naming conventions with the final decisions!!!!
--------------------------------------------------
the current tfjs library version displays a lot of warnings in the console, look out for a new tfjs version to remove them (which supports source maps)
the word "current" is used heavily within this project and should be replaced with "active"
ideally should write new code in a separate module to enable code splitting into separate chunks (check is sharing is required)
the dimensions used for recording examples were 1920x822
The export datasets button is not disabled on load time to enable the user explore the formats available
encode all non-alphanumeric symbols in index into entity code
Decided not to have a continue to edit polygon button on the popup as it made it look vulgar and does is not a functionality that users
would really want from the get-go as the rest of the polygon edit tools would enable them to ammend their polygons post creation
Decided not to use image segmentation because it works on pixel bases, where each and every boundary is drawn by an array of pixels
However, can be achieved by generating an algorithm that will draw polygons based on the boundaries found on the pixels variable's value
on line 318, file called as segment-annotator.js.
https://github.com/kyamagu/js-segment-annotator
The following repo does use fabricjs, but to only draw a line;
https://github.com/AKSHAYUBHAT/ImageSegmentation
The nice to haves:
when a user tries to drop an image on top of the app screen, it should automatically trigger uplaod (or in the case of uploading datasets - add to the mdodal)
https://webpack.js.org/guides/caching/#extracting-boilerplate
split up the result bundle into chunks to increase download
Use the following to identify the browser:
https://github.com/WhichBrowser/Parser-JavaScript?fbclid=IwAR0pxGzSJEdjUKMkWRryUI8N5c81h5YyiK1fVXCK_IamftA_vCpD21DygyM
select multiple images or shapes to delete (via ctrl)
can update the image list thumbnails by taking screenshot from farbicjs canvas when something changes
the label should not track off the screen
polygon size reduction when zoomed in is a little bit off
I believe the bounding box drawing bottom edge is not working in horizontally narrow images going by the trend
control the darkness of the background
control the thickness of shape corner sizes
control the size of labels
change label colors
rebind keys
export custom settings
animations for image upload, new label create
should alert the user when unsaved changes before leaving (potentially too intrusive for new people)
# For more information see: https://help.github.com/actions/language-and-framework-guides/publishing-nodejs-packages