Amazon Rainforest dataset for semantic segmentation V2

This database contains images used for training a fully convolutional neural network for the semantic segmentation of forested areas in images from the Sentinel-2 Level 2A Satellite.

The images refer to the RGB composition (bands 4, 3 and 2). The histogram for each image was selected as follows:

• Band 4 (red): values ranging from 103 to 2724;
• Band 3 (green): values ranging from 194 to 2888;
• Band 2 (blue): values ranging from 99 to 2798.

After that, each band was converted to a byte type (0-255).

The images are still divided into three main sets: training, validation and testing:

1. Training dataset: it contains 1.123 GeoTIFF images with 512x512 pixels and associated PNG masks (clouds indicated in white and background in black color).
2. Validation dataset: it contains 100 GeoTIFF images with 512x512 pixels and associated PNG masks used for validation step.
3. Test dataset: it contains 100 GeoTIFF images 512x512 pixels for testing.