Enhancing Model Accuracy through Advanced Transfer Learning Technique #13
Conversation
-The current implementation utilizes EfficientNet for terrain classification. - Data Agumentation includes resizing, horizontal flipping, color jitter, and random rotations. - I have loaded a pre-trained EfficientNet B0 model and modified the classifier layer to match the number of output classes (5 for different terrain types). - I noticed at high learning rates the accuracy was highly volatile so it was lowered - I defined the loss function (CrossEntropyLoss) and optimizer (Adam) with a lower learning rate for fine-tuning. - Learning rate scheduling : ReduceLROnPlateau - Included code to load model and test it using a example image - The hyper - parameter tuning was performed using gpu , but if not available code will automatically use cpu
-The current implementation utilizes EfficientNet for terrain classification. Data Agumentation includes resizing, horizontal flipping, color jitter, and random rotations. I have loaded a pre-trained EfficientNet B0 model and modified the classifier layer to match the number of output classes (5 for different terrain types). I noticed at high learning rates the accuracy was highly volatile so it was lowered I defined the loss function (CrossEntropyLoss) and optimizer (Adam) with a lower learning rate for fine-tuning. Learning rate scheduling : ReduceLROnPlateau Included code to load model and test it using a example image The hyper - parameter tuning was performed using gpu , but if not available code will automatically use cpu - added readme.md - added improvement.md
Akasxh
added
documentation
|
@Akasxh it seems the corrects labels are hactoberfest , gssoc-ext , level 3 , can you please update the labels so that the associated points are awarded. |
|
@Akasxh despite of many times telling you have not changed the labels. |
|
@AKSHITHA-CHILUKA I am not available 24/7, I have updated the labels as soon I got back to my machine. |
|
so now you are available so I will see now whether you have added all the required labels or not |
|
I have updated all the labels please check if there's any issue. |
|
|
|
Should I be assigning once it's completed or just as a issue is raised? The points are taken from prs right? |
|
both issue and pr if and only if both are labeled , then only the points will be reflected ,once you assign them you have to add all the labels including level labels . |
|
Okay done |
|
ok for further doubts please connect with @Meetjain1 |
Issue : #8
Darsh Agrawal , GSSoC 2024 extd contributor