clip-fine-tuned-satellite

This model is a fine-tuned version of openai/clip-vit-base-patch32 on the blanchon/UC_Merced dataset.
It achieves the following results on the test set:
-Accuracy: 96.9%
The original CLIP model achieves 58.8% of accuracy.

Model description

The model is a fine-tuned version of CLIP.
30% of the parameters were retrained to achieve a significant increase in accuracy after only 2 epochs.

Intended uses & limitations

The model is to be used to classify satellite images.
It was trained on the UC_Merced dataset that comprises 21 classes: agricultural, airplane, baseballdiamond, beach, buildings, chaparral, denseresidential, forest, freeway, golfcourse, harbor, intersection, mediumresidential, mobilehomepark, overpass, parkinglot, river, runway, sparseresidential, storagetanks, tenniscourt

To see how to use it, refer to the CLIP documentation or check the app using this model:
https://huggingface.co/spaces/NemesisAlm/clip-satellite-demo

Training and evaluation data

30% of the parameters trained.
Evaluated against a test set of 420 images.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 3e-05
• train_batch_size: 64
• eval_batch_size: 64
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 2
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
1.4974	1.0	20	3.0190
1.3733	2.0	40	2.9588

Framework versions

• Transformers 4.44.0
• Pytorch 2.4.0+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1