Open robinsonkwame opened 4 years ago
note: I have a copy of Sketch that I can use to design a set of clickable wire frame
@robinsonkwame should I draw the sketch and post it here?
Hmm ... let's first just list out what features a minimal product (1 suggests guiding principals that we can keep in mind but not required because we're not a full design team, etc.). We/I can sketchagainst that and run it by Eglash. So, I think minimal features for an app are:
Any thoughts or anything missing? Once we're in agreement I'll check with Eglash and then we can design a quick wireframe, etc.
@robinsonkwame that sounds fantastic. Location would help a lot with improving artisanal economy because it will help future tourists to visit the location easily. I cant think of anything else that shield be added now. This sounds great to me
great, okay, so I'm in classes all day today and tmmrw but what I'm going going do is a) propose these to Eglash, b) as mentioned in Slack, re: paper, I need to generate a mosaic of all of the swatches (unless you want to) as a figure and then I think the paper will be ready to go for Eglash to look at and we try to get it publish.
On Thu, Jun 4, 2020 at 3:09 PM sansitha nandakumar notifications@github.com wrote:
@robinsonkwame https://github.com/robinsonkwame that sounds fantastic. Location would help a lot with improving artisanal economy because it will help future tourists to visit the location easily. I cant think of anything else that shield be added now. This sounds great to me
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/robinsonkwame/authente-kente-app/issues/1#issuecomment-639060005, or unsubscribe https://github.com/notifications/unsubscribe-auth/AB5ESR4RASQRWPVUKSJH4MDRU7WM3ANCNFSM4M4GSTWA .
I will get into generating the mosaic of all swatches
cool, this library may or may not be useful/needed https://github.com/dvdtho/python-photo-mosaic, it would be useful if we could somehow mark which group of swatches were real, fake, although I bet two seperate mosaics could be generated and that would be enough
On Thu, Jun 4, 2020 at 3:18 PM sansitha nandakumar notifications@github.com wrote:
I will get into generating the mosaic of all swatches
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.
@sansi95
re: mock application, In a recent email Eglash said,
I am guessing the next steps go something like this:
1. Make a version that can run online. It would just take the training set of images we already have, but accept a new image to be tested, supplied by the user.
2. Modify the online version so that it is easier to supply a new set of training images. That will make it easier for folks in Ghana to experiment with this and become involved; and just to run experiments at our end.
3. Create cell phone version
4. Expand functionalities, depending on what Ghanaian collaborators want. One possibility is to create a feature that allows the buyer to see a video of the type of fabric they purchased being made. That could also be something that helps identify the name of the pattern and meaning for the buyer.
So this somewhat changes and solidifies our direction here, instead of a mobile application, at the moment, we need from the ML side of things:
a) an online learning version of the model; b) accessible via an API
You've looked into (b) so now we need to see what online learning solutions from (a) are available within (b). I think Tensorflow allows batched retraininable learning, although I'm not sure about convergence, which might be a solution. I think sklearn has online learning algorithms too, I have a vague sense of what's out there but I'd like to see what you find too.
Separately, later on (or at the same time, time depending) port either (a) or the static model we have now for use within a mobile app.
EDIT - also, technically, I believe @ahunnii is going to handle/help strongly with coding up (b)
User Story
As a potential CST application user I would like verify that the initial Authente-Kente application provides authentication (real or fake decisions) decisions through an interface that is both easy to understand and also leaves room for future work features.
This is weakly dependent on the type of app we're making this in, since a webapp, native app will have an UI and a web API will just have a photo upload button. The web API represents the worse case.
This requires that the
oc-svm
,kente-cloth-authentication
issues are completed.(TODO: Add Andrew and Sansitha)
Description
With a decent looking set of screens that represent both the UI flow and proposed first stage authentication functionality of the application we can verify with Dr Eglash (or anyone else) that we have an easy to use means of authenticating kente swatches (e.g. face validity).
As is common in user feedback sessions we will probably get additional feature requests and features; this feedback can be recorded, tabled for later stage future work until the first version of the application is finished.
Acceptance Requirements
Files affected
N/A