This model is a fine-tuned version of facebook/convnextv2-nano-22k-384 on the pszemraj/boulderspot dataset. It achieves the following results on the evaluation set:

Loss: 0.0340
Accuracy: 0.9883
F1: 0.9883
Precision: 0.9883
Recall: 0.9883
Matthews Correlation: 0.8962

example usage

import requests
from PIL import Image
from transformers import pipeline

pipe = pipeline(
    "image-classification",
    model="pszemraj/convnextv2-nano-22k-384-boulderspot",
)

url = "https://huggingface.co/pszemraj/convnextv2-nano-22k-384-boulderspot/resolve/main/test_img_magic_wood.png?download=true"
image = Image.open(requests.get(url, stream=True).raw)
result = pipe(image)[0]
print(result)
# image.show()

Intended uses & limitations

Classification of aerial/satellite imagery, ideally with spacial resolution 10-25 cm (i.e. for 10 cm, each pixel in the image corresonds to approx. 10 cm x 10 cm area on the ground). It may be suitable outside of that, but should be validated as other resolutions were not present in the training data.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 2e-05
train_batch_size: 16
eval_batch_size: 16
seed: 7890
gradient_accumulation_steps: 4
total_train_batch_size: 64
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: cosine
lr_scheduler_warmup_ratio: 0.05
num_epochs: 5.0

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1	Precision	Recall	Matthews Correlation
0.1102	1.0	203	0.0431	0.9839	0.9840	0.9841	0.9839	0.8590
0.0559	2.0	406	0.0476	0.9839	0.9845	0.9858	0.9839	0.8709
0.0402	3.0	609	0.0464	0.9810	0.9817	0.9831	0.9810	0.8468
0.0334	4.0	813	0.0348	0.9868	0.9869	0.9870	0.9868	0.8846
0.0445	4.99	1015	0.0340	0.9883	0.9883	0.9883	0.9883	0.8962

Framework versions

Transformers 4.39.2
Pytorch 2.4.0.dev20240328+cu121
Datasets 2.18.0
Tokenizers 0.15.2