def __getitem__(self, idx):
        group = self.id_groups.iloc[idx]
        tif_paths = group['tifPath']  # List of tif paths for this ID
        target = group['Target']
        images = []
        for tif_path in tif_paths:
            image = process_window(tif_path)  # Out  : (12, H, W)
            # Normalized RGB
            rgb_indices = [3, 2, 1, 8, 10, 11]
            bands = []
            for i, idx_band in enumerate(rgb_indices):
                band = image[idx_band]
                bands.append(
                    normalize_band(
                        band,
                        method=self.normalization_method,
                        band_index=idx_band+1,
                        sentinel2_stats=self.sentinel2_stats,
                        scale_max=self.scale_max,
                    )
                )
            img = np.stack(bands, axis=0)  # (6, H, W)
            # Compute and concatenate vegetation indices
            indices = compute_indices(image)  # (10, H, W)
            img = np.concatenate([img, indices], axis=0)  # (16, H, W)
            if isinstance(img, np.ma.MaskedArray):
                img = img.filled(0)
            img = np.moveaxis(img, 0, -1)  # (H, W, 16)  # keep for albumentations
            img = img.astype(np.float32)
            if self.transform:
                img = self.transform(image=img)["image"]
            img = torch.tensor(img, dtype=torch.float32)
            images.append(img)
        # keep only the first 7 months    
        images = images[:7]
        
        images = torch.stack(images, dim=0)  # (N_imgs, C, H, W)
        if self.to_train:
            target = rev_mapper[target]
            return images, target
        else:
            return images