mit-b0-Image_segmentation-Carvana_Image_Masking

This model is a fine-tuned version of nvidia/mit-b0.

It achieves the following results on the evaluation set:

Loss: 0.0070
Mean Iou: 0.9917
Mean Accuracy: 0.9962
Overall Accuracy: 0.9972
Per Category Iou
- Segment 0: 0.9964996655500316
- Segment 1: 0.9868763925617403
Per Category Accuracy
- Segment 0: 0.9980006976075766
- Segment 1: 0.994318466698934

Model description

For more information on how it was created, check out the following link: https://github.com/DunnBC22/Vision_Audio_and_Multimodal_Projects/blob/main/Computer%20Vision/Image%20Segmentation/Carvana%20Image%20Masking/Carvana%20Image%20Masking%20-%20Image%20Segmentation%20with%20LoRA.ipynb

Intended uses & limitations

I used this to improve my skillset. I thank all of authors of the different technologies and dataset(s) for their contributions that have made this possible.

Please make sure to properly cite the authors of the different technologies and dataset(s) as they absolutely deserve credit for their contributions.

Training and evaluation data

Dataset Source: https://www.kaggle.com/datasets/ipythonx/carvana-image-masking-png

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 0.0005
train_batch_size: 8
eval_batch_size: 8
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
num_epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss	Mean Iou	Mean Accuracy	Overall Accuracy	Segment 0 Per Category Iou	Segment 1 Per Category Iou	Segment 0 Per Category Accuracy	Segment 1 Per Category Accuracy
0.0137	1.0	509	0.0113	0.9873	0.9942	0.9957	0.9946	0.9799	0.9969	0.9915
0.011	2.0	1018	0.0096	0.9889	0.9948	0.9963	0.9953	0.9826	0.9974	0.9922
0.0096	3.0	1527	0.0087	0.9899	0.9950	0.9966	0.9958	0.9841	0.9978	0.9922
0.0089	4.0	2036	0.0082	0.9904	0.9958	0.9968	0.9959	0.9848	0.9975	0.9941
0.0086	5.0	2545	0.0078	0.9907	0.9962	0.9969	0.9961	0.9853	0.9974	0.9951
0.0082	6.0	3054	0.0077	0.9908	0.9964	0.9969	0.9961	0.9855	0.9973	0.9956
0.0081	7.0	3563	0.0072	0.9914	0.9961	0.9971	0.9964	0.9864	0.9979	0.9944
0.0081	8.0	4072	0.0071	0.9915	0.9961	0.9972	0.9964	0.9866	0.9980	0.9942
0.0089	9.0	4581	0.0070	0.9916	0.9961	0.9972	0.9965	0.9868	0.9980	0.9941
0.0076	10.0	5090	0.0070	0.9917	0.9962	0.9972	0.9965	0.9869	0.9980	0.9943

All values in the chart above are rounded to the nearest ten-thousandth.

Framework versions

Transformers 4.29.1
Pytorch 2.0.1
Datasets 2.13.1
Tokenizers 0.13.3