imagemarkup / isic-archive Goto Github PK
View Code? Open in Web Editor NEWInternational Skin Imaging Collaboration: Melanoma Project
Home Page: https://isic-archive.com/
License: Apache License 2.0
International Skin Imaging Collaboration: Melanoma Project
Home Page: https://isic-archive.com/
License: Apache License 2.0
Manual annotation of a lesion boundary sometimes leads to the disappearance of the lesion window. This occurs at inconsistent times. Temporary solution is to refresh page in chrome.
Priority: Need more test scenarios
When submitting an image in phase 1b the image does move on to phase 1c (you can count it in the dashboard). The problem is the submit button appears to not work since the user isn't notified of any changes and isn't brought back to the dashboard.
Changes made in phase 1b to lesion boundary and superpixel size are not saved when opening up an image in phase 1c.
Also hide "Collections" tab on the left.
Since users are now private, group admins that are not site admins cannot see the list of users.
For example, it would show:
in addition to the current layout:
This could be useful if we start getting a lot of different image datasets / folders.
Phase 2 is missing a few lesion level selections as per Dr. Marchetti.
During review process, a user can select boxes on the screen, this should be disabled if possible (low priority as it actually doesn’t get saved).
This behavior seems to only be affecting adjustments to superpixel size.
Complexity: Medium to low
Solution: ADD in logic so that IF new boundaries exist in current Phase use it, otherwise use one from previous phase--- current logic most likely just grabs the one from the last phase
Still make user list available to admins and "study creators".
I just wanted to drop you a note to say that this is a really impressive project. A colleague of mine stumbled across the archive when reading about the upcoming ISBI challenge.
Anyhow, I wanted to make you aware that I have created a little Julia package to access the images conveniently from Julia via the REST API. We are interested to investigate deep learning approaches on the data, so I thought others might find this useful as well. I hope you don't mind.
Let me know if you want me to change anything or remove it from Github.
The zoom used for image annotation in phase 2 is too high for images of small lesions causing pixellation and loss of detail presented to the user.
First the conda module must be installed. This might be in latest trunk but my ansible 1.8.4 did not have it so i downloaded https://github.com/synthicity/ansible-conda and copied it to my modules/extras/pagkaging dir.
Second, the playbook failed because ansible_user_dir was undefined. Added
vars:
ansible_user_dir: /home/vagrant
to vagrant-playbook.yml
Third, I had to increase the default vm memory.
Running "shell:readServerConfig" (shell) task
Warning: spawn ENOMEM Use --force to continue.
Shut down vm. Machine->Settings-> increase memory to 2048
When logging into a new browser or new computer for the first time, users get logged out automatically from ISIC. Occurs when logging in with password (not google, although haven't tested with google).
Especially tends to happen when logging in, then clicking on Image Tasks, then clicking on the Back button to go back to the ISIC homepage.
Using Chrome on Windows 7 currently.
Also, an endpoint for getting all annotation objects for a given study + given image.
Also add "Angulated Lines" as a feature.
In phase 1a if the user trashes the annotation at the review page there is no option to then redo that annotation. Temporary solution is to refresh page in chrome.
Priority: Low
Complexity: Moderate to Hard?
This needs to be documented, or an error message needs to be added if the user fails to click a seed point in the image.
Allow more than one user to be working on images from the same phase but not be annotating the exact same images.
Task complexity: Medium to Low
Solution: Add add’l checked out flag which is set when image enters the work queue
Potential issues: Need to every once in a while “clear” the flag in case someone stops working on an image and doesn’t submit it
ADD TO Metadata model:
image_checked_out = Boolean
image_checkout_time = TIMESTAMP #So can set cron job to put images back in the work pool
Allow users to pick which phase they would like to work on/annotate.
Complexity: Moderate to Hard
The thumbnail-friendly .tif
image is available, but we should also have the original full-quality .jpg
image.
Datasets with this problem:
uda_10-9-2014_560imgs
Have the lesion window and all menu bars on the screen for all common resolutions. Optimal screen resolution at this moment looks to be 1920×1080 and up.
Complexity: Medium
Time to implement: High
Solution: Buy people nicer laptops
When auto selecting the lesion boundary of a particularly small sized image, the user needs to zoom in and out iteratively before the boundary can be set.
Issue: For images < CANVAS size/width, the image boundaries are likely not being populated properly, so we are basing bogus coordinates to the segmenter…
Current work around: Zoom in/Zoom out and the boundaries seem to be set
Issue: Figuring out where in the code these parameters are being pulled from, and fix it
Complexity: Medium
Add an error message informing the user that a certain phase does not have any more images for annotation.
Priority: High
Solution: Add in graceful message if QUEUE == EMPTY
Long Term Solution: #2
Either make all images larger, or add a preview window.
The uda0000040319997d1001 image that loads in the uda_10-9-2014_560imgs folder in phase 1c is corrupt. It cannot be annotated and is blocking progress for submission into phase 2, annotation by Dr. Marghoob, and completion of the UDA1 study.
Possible schemes:
ISIC_{contributor-id}_{batch-id}
{contributor-id}_{batch-id}
For example:
ISIC_anon3_5
ISIC_SONIC_br-bros
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google ❤️ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.