minor update

Finalised everything and trained using torchrun for kaggle 2*t4 optimization

Its now completely trained for longitudnal reporting though it uses a lot gpu resourses I somehow rained it on 2*t4’s over a long time its not all there will be many future updates and my final aim is still to make AetherMed something that helps atleast 1 radiolgist every single day. with that said the journey will continue for sure.

The future versions are gonna have :-

• Much better ui , auth and database features
• PDF BASED DETAILED REPORT GENERATION with Grad-Cam.
• Even better AI capabilities
• and many other features

tweaked ui and wrote code for || training

okay so i used a kicad plugin called pcb2blender so i can export my kicad pcb to blender without having to add materials and stuff myself which makes the workflow sooo much smoother, after i imported them into blender i just setup camera with some point and area lights are just made these pretty renders with eevee!!!

Training

So, basically I on an old hp probook with 4gb ram ltrelly and I am treid to train it on colab but it doesn’t haves enough vram so now I am going for kaggle as it have 2*t4. I also wrote a lot of code past hrs now we are really near to begin training maybe even in 2 weeks We may see V2 live. let hope for the best. Peace

Model loading and training - Part 1

So, basically past few days weren’t went as expected. But finally toady i see peices meeting together.I have setted up parrlel downloading while traing by dedication few threads to it and i hopes to make my training pipline so stong so it only remains as hard as changing variables to choose between which models to fine tune and d how much data to use 31st will be my final deadline for launching v1 then in future updates i will add auth and database(scans storage,chats stored etc) i will also try to meet some radiolgists and have them test AetherMEd so lets see how following days goes.

So i finally completed all the DATA PREPRCESSING it was probably the most time consuming part but it was well worth it as shaping the longitudnal data is also an hard task i also wrote code for parrally downloading and deleting images on 4 threads whil model trains on others.

1)Chapter 1 - STARTING
2)Getting our ui ready
3)Setting up shadcn, next.js app and wakatime with webstorm
4)Fixed Next js compiler issue and made Navbar
5)Completed Most of the landing Page
6)setted up dashboard (maybe)
7)setted up conda env
8)EDA
9)data preprocessing