search

Akasxh / Terrain-Recognition

High accuracy, explainable, lightweight CNN for terrain recognition.
GNU General Public License v3.0
6 stars 9 forks source link

Enhancing Model Accuracy through Advanced Transfer Learning Technique #8

Open DarshAgrawal14 opened 1 day ago

DarshAgrawal14 commented 1 day ago

Current Validation Accuracy: 91.93% Current Test Accuracy: 92.40%

Teja-m9 commented 1 day ago

hey @DarshAgrawal14 .Please assign this issue to me and i have some experienece in doing these tasks. I am pretty much sure that i can solve this problem in less amount of time

DarshAgrawal14 commented 1 day ago

this issue is just my approach for a issue given by the owner of the repo, If you are interested then you checkout Issue : #1 .

Akasxh commented 1 day ago

I just reviewed, it seems fantastic. To make it full fledged could you make a separate readme along with eff net model architecture and explain how its working in brief. -brief working -architectural diagram -how heavy the model is(space and inference speed) -also create a seperate heading in improvement.md on how to further improve it Could you add all this as well in your pull req

Thank you.

Akasxh commented 1 day ago

@Teja-m9 you can make a separate issue on his approach if you can improve the work.

DarshAgrawal14 commented 22 hours ago

got it, so do you want me add the readme and improvement md in this issue only or should i create a separate one? and should i add the data regarding transfer learning in a new directory , so that it doesn't conflict with available versions of readme and improvement md?

Akasxh commented 22 hours ago

Please do it in this issue itself.

Also do make a seperate readme for it but in the original main readme, make a link to your readme as a hypertext that there's a model with this much accuracy, size and technique where other technique can be added in future.

DarshAgrawal14 commented 15 hours ago

@Akasxh i have added the mentioned files and changes