--- license: apache-2.0 task_categories: - question-answering language: - en tags: - remote_sensing - vlm size_categories: - 10K- Eardial_downstream_task_datasets/Classification/AID/test/data-00000-of-00001.arrow - split: UCM_sample path: >- Eardial_downstream_task_datasets/Classification/UCM/data-00000-of-00001.arrow - split: BigEarthNet_RGB_sample path: >- Eardial_downstream_task_datasets/Classification/BigEarthNet_RGB/BigEarthNet_test/data-00000-of-00004.arrow - split: WHU_sample path: >- Eardial_downstream_task_datasets/Classification/WHU_19/data-00000-of-00001.arrow - config_name: GeoChat_Bench data_files: - split: GeoChat_Bench_sample path: >- Eardial_downstream_task_datasets/Detection/Geochat_Bench/data-00000-of-00001.arrow - config_name: Detection data_files: - split: NWPU_VHR_10_sample path: >- Eardial_downstream_task_datasets/Detection/NWPU_VHR_10_test/data-00000-of-00001.arrow - split: Swimming_pool_dataset_sample path: >- Eardial_downstream_task_datasets/Detection/Swimming_pool_dataset_test/data-00000-of-00001.arrow - split: ship_dataset_v0_sample path: >- Eardial_downstream_task_datasets/Detection/ship_dataset_v0_test/data-00000-of-00001.arrow - split: urban_tree_crown_sample path: >- Eardial_downstream_task_datasets/Detection/urban_tree_crown_detection_test/data-00000-of-00001.arrow - config_name: Region_captioning data_files: - split: NWPU_VHR_10 path: >- Eardial_downstream_task_datasets/Region_captioning/NWPU_VHR_10_test_region_captioning/data-00000-of-00001.arrow - config_name: Image_captioning data_files: - split: sydney_Captions path: >- Eardial_downstream_task_datasets/Image_captioning/sydney_Captions/sydney_Captions_test/data-00000-of-00001.arrow - split: UCM_Captions path: >- Eardial_downstream_task_datasets/Image_captioning/UCM_Captions/UCM_Captions_test/data-00000-of-00001.arrow - split: RSICD_Captions path: >- Eardial_downstream_task_datasets/Image_captioning/RSICD_Captions/RSICD_Captions_test/data-00000-of-00001.arrow --- # 🌍 EarthDial-Dataset The **EarthDial-Dataset** is a curated collection of evaluation-only datasets focused on remote sensing and Earth observation downstream tasks. It is designed to benchmark **vision-language models (VLMs)** and **multimodal reasoning systems** on real-world scenarios involving satellite and aerial imagery. --- ## 📚 Key Features - **Evaluation-focused**: All datasets are for inference/testing only — no train/val splits. - **Diverse Tasks**: - Classification - Object Detection - Change Detection - Grounding Description - Region Captioning - Image Captioning - Visual Question Answering (GeoChat Bench) - **Remote Sensing Specific**: Tailored for multispectral, RGB, and high-resolution satellite data. - **Multimodal Format**: Includes images, questions, captions, annotations, and geospatial metadata. --- ## 🗂️ Dataset Structure The dataset is structured under the root folder: `EarthDial_downstream_task_datasets/` Each task has its own subdirectory with `.arrow`-formatted shards, structured as: ```bash EarthDial_downstream_task_datasets/ │ ├── Classification/ │ ├── AID/ │ │ └── test/data-00000-of-00001.arrow │ └── ... │ ├── Detection/ │ ├── NWPU_VHR_10_test/ │ ├── Swimming_pool_dataset_test/ │ └── ... │ ├── Region_captioning/ │ └── NWPU_VHR_10_test_region_captioning/ │ ├── Image_captioning/ │ ├── RSICD_Captions/ │ └── UCM_Captions/ │... ## 🗂️ Example data usage from datasets import load_dataset dataset = load_dataset( "akshaydudhane/EarthDial-Dataset", data_dir="EarthDial_downstream_task_datasets/Classification/AID/test" ) ## Example Demo Usage import argparse import torch from PIL import Image from transformers import AutoTokenizer from earthdial.model.internvl_chat import InternVLChatModel from earthdial.train.dataset import build_transform def run_single_inference(args): # Load model and tokenizer from Hugging Face Hub print(f"Loading model and tokenizer from Hugging Face: {args.checkpoint}") tokenizer = AutoTokenizer.from_pretrained(args.checkpoint, trust_remote_code=True, use_fast=False) model = InternVLChatModel.from_pretrained( args.checkpoint, low_cpu_mem_usage=True, torch_dtype=torch.bfloat16, device_map="auto" if args.auto else None, load_in_8bit=args.load_in_8bit, load_in_4bit=args.load_in_4bit ).eval() if not args.load_in_8bit and not args.load_in_4bit and not args.auto: model = model.cuda() # Load and preprocess image image = Image.open(args.image_path).convert("RGB") image_size = model.config.force_image_size or model.config.vision_config.image_size transform = build_transform(is_train=False, input_size=image_size, normalize_type='imagenet') pixel_values = transform(image).unsqueeze(0).cuda().to(torch.bfloat16) # Generate answer generation_config = { "num_beams": args.num_beams, "max_new_tokens": 100, "min_new_tokens": 1, "do_sample": args.temperature > 0, "temperature": args.temperature, } answer = model.chat( tokenizer=tokenizer, pixel_values=pixel_values, question=args.question, generation_config=generation_config, verbose=True ) print("\n=== Inference Result ===") print(f"Question: {args.question}") print(f"Answer: {answer}") if __name__ == "__main__": parser = argparse.ArgumentParser() parser.add_argument('--checkpoint', type=str, required=True, help='Model repo ID on Hugging Face Hub') parser.add_argument('--image-path', type=str, required=True, help='Path to local input image') parser.add_argument('--question', type=str, required=True, help='Question to ask about the image') parser.add_argument('--num-beams', type=int, default=5) parser.add_argument('--temperature', type=float, default=0.0) parser.add_argument('--load-in-8bit', action='store_true') parser.add_argument('--load-in-4bit', action='store_true') parser.add_argument('--auto', action='store_true') args = parser.parse_args() run_single_inference(args) python demo_infer.py \ --checkpoint akshaydudhane/EarthDial_4B_RGB \ --image-path ./test.jpg \ --question "Which road has more vehicles?" \ --auto