swin-tiny-patch4-window7-224_ft_mango_leaf_disease

This model was trained from scratch on the imagefolder dataset. It achieves the following results on the evaluation set:

• Loss: 0.0089
• Accuracy: 0.9986

Model description

Multiclass image classification model based on swin-tiny-patch4-window7-224 and fine-tuned with Mango🥭 Leaf🍃🍂 Disease Dataset. Model was trained on 8 classes based on mango leaves health : Anthracnose, Bacterial Canker, Cutting Weevil, Die Back, Gall Midge, Powdery Mildew, Sooty Mould, Healthy

Intended uses & limitations

More information needed

Training and evaluation data

Traning and evaluation data are from this Kaggle dataset Mango🥭 Leaf🍃🍂 Disease Dataset. Amount of images used was 90% of total images (3600 of 4000, 450 images from each class).

Training procedure

Dataset split : 75% train set, 20% validation set, 5% test set.

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 64
• eval_batch_size: 64
• seed: 143
• gradient_accumulation_steps: 4
• total_train_batch_size: 256
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
No log	0.93	10	0.1208	0.9931
0.1082	1.95	21	0.0551	0.9958
0.1082	2.98	32	0.0297	0.9958
0.0342	4.0	43	0.0189	0.9986
0.0342	4.93	53	0.0156	0.9972
0.0164	5.95	64	0.0122	0.9972
0.0164	6.98	75	0.0100	0.9986
0.0099	8.0	86	0.0096	0.9986
0.0099	8.93	96	0.0090	0.9986
0.0085	9.3	100	0.0089	0.9986

Framework versions

• Transformers 4.33.3
• Pytorch 2.0.1+cu118
• Datasets 2.14.5
• Tokenizers 0.13.3