import os import torch import yaml import numpy as np import gradio as gr from einops import rearrange from functools import partial from huggingface_hub import hf_hub_download # pull files from hub token = os.environ.get("HF_TOKEN", None) config_path = hf_hub_download(repo_id="ibm-nasa-geospatial/Prithvi-EO-2.0-300M-TL", filename="config.json", token=token) checkpoint = hf_hub_download(repo_id="ibm-nasa-geospatial/Prithvi-EO-2.0-300M-TL", filename='Prithvi_EO_V2_300M_TL.pt', token=token) model_def = hf_hub_download(repo_id="ibm-nasa-geospatial/Prithvi-EO-2.0-300M-TL", filename='prithvi_mae.py', token=token) model_inference = hf_hub_download(repo_id="ibm-nasa-geospatial/Prithvi-EO-2.0-300M-TL", filename='inference.py', token=token) os.system(f'cp {model_def} .') os.system(f'cp {model_inference} .') from prithvi_mae import PrithviMAE from inference import process_channel_group, _convert_np_uint8, load_example, run_model def extract_rgb_imgs(input_img, rec_img, mask_img, channels, mean, std): """ Wrapper function to save Geotiff images (original, reconstructed, masked) per timestamp. Args: input_img: input torch.Tensor with shape (C, T, H, W). rec_img: reconstructed torch.Tensor with shape (C, T, H, W). mask_img: mask torch.Tensor with shape (C, T, H, W). channels: list of indices representing RGB channels. mean: list of mean values for each band. std: list of std values for each band. output_dir: directory where to save outputs. meta_data: list of dicts with geotiff meta info. """ rgb_orig_list = [] rgb_mask_list = [] rgb_pred_list = [] for t in range(input_img.shape[1]): rgb_orig, rgb_pred = process_channel_group(orig_img=input_img[:, t, :, :], new_img=rec_img[:, t, :, :], channels=channels, mean=mean, std=std) rgb_mask = mask_img[channels, t, :, :] * rgb_orig # extract images rgb_orig_list.append(_convert_np_uint8(rgb_orig).transpose(1, 2, 0)) rgb_mask_list.append(_convert_np_uint8(rgb_mask).transpose(1, 2, 0)) rgb_pred_list.append(_convert_np_uint8(rgb_pred).transpose(1, 2, 0)) # Add white dummy image values for missing timestamps dummy = np.ones((20, 20), dtype=np.uint8) * 255 num_dummies = 4 - len(rgb_orig_list) if num_dummies: rgb_orig_list.extend([dummy] * num_dummies) rgb_mask_list.extend([dummy] * num_dummies) rgb_pred_list.extend([dummy] * num_dummies) outputs = rgb_orig_list + rgb_mask_list + rgb_pred_list return outputs def predict_on_images(data_files: list, config_path: str, checkpoint: str, mask_ratio: float = None): try: data_files = [x.name for x in data_files] print('Path extracted from example') except: print('Files submitted through UI') # Get parameters -------- print('This is the printout', data_files) with open(config_path, 'r') as f: config = yaml.safe_load(f)['pretrained_cfg'] batch_size = 8 bands = config['bands'] num_frames = len(data_files) mean = config['mean'] std = config['std'] coords_encoding = config['coords_encoding'] img_size = config['img_size'] mask_ratio = mask_ratio or config['mask_ratio'] assert num_frames <= 4, "Demo only supports up to four timestamps" if torch.cuda.is_available(): device = torch.device('cuda') else: device = torch.device('cpu') print(f"Using {device} device.\n") # Loading data --------------------------------------------------------------------------------- input_data, temporal_coords, location_coords, meta_data = load_example(file_paths=data_files, mean=mean, std=std) if len(temporal_coords) != num_frames and 'time' in coords_encoding: coords_encoding.pop('time') if not len(location_coords) and 'location' in coords_encoding: coords_encoding.pop('location') # Create model and load checkpoint ------------------------------------------------------------- config.update( num_frames=num_frames, coords_encoding=coords_encoding, ) model = PrithviMAE(**config) total_params = sum(p.numel() for p in model.parameters() if p.requires_grad) print(f"\n--> Model has {total_params:,} parameters.\n") model.to(device) state_dict = torch.load(checkpoint, map_location=device, weights_only=False) # discard fixed pos_embedding weight for k in list(state_dict.keys()): if 'pos_embed' in k: del state_dict[k] model.load_state_dict(state_dict, strict=False) print(f"Loaded checkpoint from {checkpoint}") # Running model -------------------------------------------------------------------------------- model.eval() channels = [bands.index(b) for b in ['B04', 'B03', 'B02']] # BGR -> RGB # Reflect pad if not divisible by img_size original_h, original_w = input_data.shape[-2:] pad_h = img_size - (original_h % img_size) pad_w = img_size - (original_w % img_size) input_data = np.pad(input_data, ((0, 0), (0, 0), (0, 0), (0, pad_h), (0, pad_w)), mode='reflect') # Build sliding window batch = torch.tensor(input_data, device='cpu') windows = batch.unfold(3, img_size, img_size).unfold(4, img_size, img_size) h1, w1 = windows.shape[3:5] windows = rearrange(windows, 'b c t h1 w1 h w -> (b h1 w1) c t h w', h=img_size, w=img_size) # Split into batches if number of windows > batch_size num_batches = windows.shape[0] // batch_size if windows.shape[0] > batch_size else 1 windows = torch.tensor_split(windows, num_batches, dim=0) # Run model rec_imgs = [] mask_imgs = [] for x in windows: temp_coords = torch.Tensor([temporal_coords] * len(x)) loc_coords = torch.Tensor([location_coords[0]] * len(x)) rec_img, mask_img = run_model(model, x, temp_coords, loc_coords, mask_ratio, device) rec_imgs.append(rec_img) mask_imgs.append(mask_img) rec_imgs = torch.concat(rec_imgs, dim=0) mask_imgs = torch.concat(mask_imgs, dim=0) # Build images from patches rec_imgs = rearrange(rec_imgs, '(b h1 w1) c t h w -> b c t (h1 h) (w1 w)', h=img_size, w=img_size, b=1, c=len(bands), t=num_frames, h1=h1, w1=w1) mask_imgs = rearrange(mask_imgs, '(b h1 w1) c t h w -> b c t (h1 h) (w1 w)', h=img_size, w=img_size, b=1, c=len(bands), t=num_frames, h1=h1, w1=w1) # Cut padded images back to original size rec_imgs_full = rec_imgs[..., :original_h, :original_w] mask_imgs_full = mask_imgs[..., :original_h, :original_w] batch_full = batch[..., :original_h, :original_w] # Build RGB images for d in meta_data: d.update(count=3, dtype='uint8', compress='lzw', nodata=0) outputs = extract_rgb_imgs(batch_full[0, ...], rec_imgs_full[0, ...], mask_imgs_full[0, ...], channels, mean, std) print("Done!") return outputs run_inference = partial(predict_on_images, config_path=config_path,checkpoint=checkpoint) with gr.Blocks() as demo: gr.Markdown(value='# Prithvi-EO-2.0 image reconstruction demo') gr.Markdown(value=''' Prithvi-EO-2.0 is the second generation EO foundation model developed by the IBM and NASA team. The temporal ViT is train on 4.2M Harmonised Landsat Sentinel 2 (HLS) samples with four timestamps each, using the Masked AutoEncoder learning strategy. The model includes spatial and temporal attention across multiple patches and timestamps. Additionally, temporal and location information is added to the model input via embeddings. More info about the model are available [here](https://huggingface.co/ibm-nasa-geospatial/Prithvi-EO-2.0-300M-TL).\n This demo showcases the image reconstruction over one to four timestamps. The model randomly masks out some proportion of the images and reconstructs them based on the not masked portion of the images. The reconstructed images are merged with the visible unmasked patches. We recommend submitting images of size 224 to ~1000 pixels for faster processing time. Images bigger than 224x224 are processed using a sliding window approach which can lead to artefacts between patches.\n The user needs to provide the HLS geotiff images, including the following channels in reflectance units: Blue, Green, Red, Narrow NIR, SWIR, SWIR 2. Optionally, the location information is extracted from the tif files while the temporal information can be provided in the filename in the format `T