Create APP

.gitattributes

@@ -1,35 +1,35 @@
-*.7z filter=lfs diff=lfs merge=lfs -text
-*.arrow filter=lfs diff=lfs merge=lfs -text
-*.bin filter=lfs diff=lfs merge=lfs -text
-*.bz2 filter=lfs diff=lfs merge=lfs -text
-*.ckpt filter=lfs diff=lfs merge=lfs -text
-*.ftz filter=lfs diff=lfs merge=lfs -text
-*.gz filter=lfs diff=lfs merge=lfs -text
-*.h5 filter=lfs diff=lfs merge=lfs -text
-*.joblib filter=lfs diff=lfs merge=lfs -text
-*.lfs.* filter=lfs diff=lfs merge=lfs -text
-*.mlmodel filter=lfs diff=lfs merge=lfs -text
-*.model filter=lfs diff=lfs merge=lfs -text
-*.msgpack filter=lfs diff=lfs merge=lfs -text
-*.npy filter=lfs diff=lfs merge=lfs -text
-*.npz filter=lfs diff=lfs merge=lfs -text
-*.onnx filter=lfs diff=lfs merge=lfs -text
-*.ot filter=lfs diff=lfs merge=lfs -text
-*.parquet filter=lfs diff=lfs merge=lfs -text
-*.pb filter=lfs diff=lfs merge=lfs -text
-*.pickle filter=lfs diff=lfs merge=lfs -text
-*.pkl filter=lfs diff=lfs merge=lfs -text
-*.pt filter=lfs diff=lfs merge=lfs -text
-*.pth filter=lfs diff=lfs merge=lfs -text
-*.rar filter=lfs diff=lfs merge=lfs -text
-*.safetensors filter=lfs diff=lfs merge=lfs -text
-saved_model/**/* filter=lfs diff=lfs merge=lfs -text
-*.tar.* filter=lfs diff=lfs merge=lfs -text
-*.tar filter=lfs diff=lfs merge=lfs -text
-*.tflite filter=lfs diff=lfs merge=lfs -text
-*.tgz filter=lfs diff=lfs merge=lfs -text
-*.wasm filter=lfs diff=lfs merge=lfs -text
-*.xz filter=lfs diff=lfs merge=lfs -text
-*.zip filter=lfs diff=lfs merge=lfs -text
-*.zst filter=lfs diff=lfs merge=lfs -text
-*tfevents* filter=lfs diff=lfs merge=lfs -text
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text

README.md

@@ -1,13 +1,13 @@
----
-title: Location Predictor
-emoji: 📚
-colorFrom: pink
-colorTo: red
-sdk: gradio
-sdk_version: 4.38.1
-app_file: app.py
-pinned: false
-license: mit
----
-
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+---
+title: Location Predictor
+emoji: 🌍
+colorFrom: red
+colorTo: red
+sdk: gradio
+sdk_version: 4.38.1
+app_file: app.py
+pinned: false
+license: mit
+---
+
+Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

app.py

@@ -0,0 +1,128 @@
+import torch
+from geoclip import GeoCLIP
+from PIL import Image
+import tempfile
+from pathlib import Path
+import gradio as gr
+import spaces
+from geopy.geocoders import Nominatim
+from transformers import CLIPProcessor, CLIPModel
+from torchvision import transforms
+import reverse_geocoder as rg
+from models.huggingface import Geolocalizer
+import folium
+import json
+from geopy.exc import GeocoderTimedOut
+
+if torch.cuda.is_available():
+    geoclip_model = GeoCLIP().to("cuda")
+else:
+    geoclip_model = GeoCLIP()
+
+geolocator = Nominatim(user_agent="predictGeolocforImage")
+
+streetclip_model = CLIPModel.from_pretrained("geolocal/StreetCLIP")
+streetclip_processor = CLIPProcessor.from_pretrained("geolocal/StreetCLIP")
+labels = ['Albania', 'Andorra', 'Argentina', 'Australia', 'Austria', 'Bangladesh', 'Belgium', 'Bermuda', 'Bhutan', 'Bolivia', 'Botswana', 'Brazil', 'Bulgaria', 'Cambodia', 'Canada', 'Chile', 'China', 'Colombia', 'Croatia', 'Czech Republic', 'Denmark', 'Dominican Republic', 'Ecuador', 'Estonia', 'Finland', 'France', 'Germany', 'Ghana', 'Greece', 'Greenland', 'Guam', 'Guatemala', 'Hungary', 'Iceland', 'India', 'Indonesia', 'Ireland', 'Israel', 'Italy', 'Japan', 'Jordan', 'Kenya', 'Kyrgyzstan', 'Laos', 'Latvia', 'Lesotho', 'Lithuania', 'Luxembourg', 'Macedonia', 'Madagascar', 'Malaysia', 'Malta', 'Mexico', 'Monaco', 'Mongolia', 'Montenegro', 'Netherlands', 'New Zealand', 'Nigeria', 'Norway', 'Pakistan', 'Palestine', 'Peru', 'Philippines', 'Poland', 'Portugal', 'Puerto Rico', 'Romania', 'Russia', 'Rwanda', 'Senegal', 'Serbia', 'Singapore', 'Slovakia', 'Slovenia', 'South Africa', 'South Korea', 'Spain', 'Sri Lanka', 'Swaziland', 'Sweden', 'Switzerland', 'Taiwan', 'Thailand', 'Tunisia', 'Turkey', 'Uganda', 'Ukraine', 'United Arab Emirates', 'United Kingdom', 'United States', 'Uruguay']
+
+IMAGE_SIZE = (224, 224)
+GEOLOC_MODEL_NAME = "osv5m/baseline"
+geoloc_model = Geolocalizer.from_pretrained(GEOLOC_MODEL_NAME)
+geoloc_model.eval()
+
+def transform_image(image):
+    transform = transforms.Compose([
+        transforms.Resize(IMAGE_SIZE),
+        transforms.ToTensor(),
+        transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
+    ])
+    return transform(image).unsqueeze(0)
+
+def create_map(lat, lon):
+    m = folium.Map(location=[lat, lon], zoom_start=4)
+    folium.Marker([lat, lon]).add_to(m)
+    map_html = m._repr_html_()
+    return map_html
+
+def get_country_coordinates(country_name):
+    try:
+        location = geolocator.geocode(country_name, timeout=10)
+        if location:
+            return location.latitude, location.longitude
+    except GeocoderTimedOut:
+        return None
+    return None
+
+@spaces.GPU
+def predict_geoclip(image):
+    with tempfile.TemporaryDirectory() as tmp_dir:
+        tmppath = Path(tmp_dir) / "tmp.jpg"
+        image.save(str(tmppath))
+        top_pred_gps, top_pred_prob = geoclip_model.predict(str(tmppath), top_k=50)
+        
+    predictions = []
+    for i in range(1):
+        lat, lon = top_pred_gps[i]
+        probpercent = top_pred_prob[i] * 100
+        location = geolocator.reverse((lat, lon), exactly_one=True)
+        address = location.raw['address']
+        city = address.get('city', '')
+        country = address.get('country', '')
+        prediction = f"Latitude: {lat:.6f}, Longitude: {lon:.6f} - Country: {country}"
+        predictions.append(prediction)
+        
+    map_html = create_map(lat, lon)
+    return "\n".join(predictions), map_html
+
+@spaces.GPU
+def classify_streetclip(image):
+    inputs = streetclip_processor(text=labels, images=image, return_tensors="pt", padding=True)
+    with torch.no_grad():
+        outputs = streetclip_model(**inputs)
+    logits_per_image = outputs.logits_per_image
+    prediction = logits_per_image.softmax(dim=1)
+    confidences = {labels[i]: float(prediction[0][i].item()) for i in range(len(labels))}
+    
+    sorted_confidences = sorted(confidences.items(), key=lambda item: item[1], reverse=True)
+    top_label, top_confidence = sorted_confidences[0]
+    coords = get_country_coordinates(top_label)
+    map_html = create_map(*coords) if coords else "Map not available"
+    return f"Country: {top_label}", map_html
+
+def infer(image):
+    try:
+        img_tensor = transform_image(image)
+        gps_radians = geoloc_model(img_tensor)
+        gps_degrees = torch.rad2deg(gps_radians).squeeze(0).cpu().tolist()
+        lat, lon = gps_degrees[0], gps_degrees[1]
+        location_query = rg.search((lat, lon))[0]
+        location_name = f"{location_query['name']}, {location_query['admin1']}, {location_query['cc']}"
+        map_html = create_map(lat, lon)
+        return f"Latitude: {lat:.6f}, Longitude: {lon:.6f} - Country: {location_query['admin1']} - {location_query['cc']}", map_html
+    except Exception as e:
+        return f"Failed to predict the location: {e}", None
+
+geoclip_interface = gr.Interface(
+    fn=predict_geoclip,
+    inputs=gr.Image(type="pil", label="Upload Image", elem_id="geoclip_image_input"),
+    outputs=[gr.Textbox(label="Prediction", elem_id="geoclip_output"), gr.HTML(label="Map", elem_id="geoclip_map_output")],
+    title="GeoCLIP"
+)
+
+streetclip_interface = gr.Interface(
+    fn=classify_streetclip,
+    inputs=gr.Image(type="pil", label="Upload Image", elem_id="streetclip_image_input"),
+    outputs=[gr.Textbox(label="Prediction", elem_id="streetclip_output"), gr.HTML(label="Map", elem_id="streetclip_map_output")],
+    title="StreetCLIP"
+)
+
+osv5m_interface = gr.Interface(
+    fn=infer,
+    inputs=gr.Image(label="Upload Image", type="pil", elem_id="osv5m_image_input"),
+    outputs=[gr.Textbox(label="Prediction", elem_id="result_text"), gr.HTML(label="Map", elem_id="map_output")],
+    title="OSV-5M Baseline"
+)
+
+demo = gr.TabbedInterface([geoclip_interface, streetclip_interface, osv5m_interface], tab_names=["GeoCLIP", "StreetCLIP", "OSV-5M Baseline"])
+
+demo.launch()

configs/computer/a100.yaml

@@ -0,0 +1,8 @@
+devices: 1
+progress_bar_refresh_rate: 2
+num_workers: 8
+sync_batchnorm: False
+accelerator: gpu
+precision: 32
+strategy: auto
+num_nodes: 1

configs/computer/cluster-node-a100.yaml

@@ -0,0 +1,8 @@
+devices: 8
+num_workers: 8
+progress_bar_refresh_rate: 2
+sync_batchnorm: True
+accelerator: gpu
+precision: 32
+strategy: ddp
+num_nodes: 1

configs/computer/cluster-node-v100.yaml

@@ -0,0 +1,8 @@
+devices: 4
+num_workers: 10
+progress_bar_refresh_rate: 2
+sync_batchnorm: True
+accelerator: gpu
+precision: 32
+strategy: ddp
+num_nodes: 1

configs/computer/cpu.yaml

@@ -0,0 +1,8 @@
+devices: null
+num_workers: 0
+progress_bar_refresh_rate: 2
+sync_batchnorm: False
+accelerator: cpu
+precision: 32
+strategy: auto
+num_nodes: null

configs/computer/v100.yaml

@@ -0,0 +1,8 @@
+devices: 1
+num_workers: 10
+progress_bar_refresh_rate: 2
+sync_batchnorm: False
+accelerator: gpu
+precision: 32
+strategy: auto
+num_nodes: 1

configs/config.yaml

@@ -0,0 +1,89 @@
+defaults:
+  - model: default
+  - computer: v100
+  - dataset: osv5m
+  - _self_
+  - exp: ???
+
+model:
+  val_metrics:
+    _target_: metrics.distance_based.HaversineMetrics
+    acc_radiuses:
+      - 1
+      - 25
+      - 200
+      - 750
+      - 2500
+    acc_area: []
+    aux_data: ${aux_data}
+  test_metrics:
+    _target_: metrics.distance_based.HaversineMetrics
+    acc_radiuses:
+      - 1
+      - 25
+      - 200
+      - 750
+      - 2500
+    acc_area: ${areas}
+    aux_data: ${aux_data}
+
+datamodule:
+  _target_: data.datamodule.ImageDataModule
+  train_dataset: ${dataset.train_dataset}
+  val_dataset: ${dataset.val_dataset}
+  test_dataset: ${dataset.test_dataset}
+  global_batch_size: ${dataset.global_batch_size}
+  num_workers: ${computer.num_workers}
+  num_nodes: ${computer.num_nodes}
+  num_devices: ${computer.devices}
+  val_proportion: 0.1
+
+trainer:
+  _target_: pytorch_lightning.Trainer
+  devices: ${computer.devices}
+  accelerator: ${computer.accelerator}
+  strategy: ${computer.strategy}
+  num_nodes: ${computer.num_nodes}
+  precision: ${computer.precision}
+  max_epochs: ${max_epochs}
+
+logger:
+  _target_: pytorch_lightning.loggers.WandbLogger
+  save_dir: ${root_dir}
+  name: ${experiment_name}
+  project: plonk
+  log_model: False
+  offline: False
+  entity: imaginelab
+
+checkpoints:
+  _target_: pytorch_lightning.callbacks.ModelCheckpoint
+  dirpath: ${root_dir}/checkpoints/${experiment_name}
+  filename: 'epoch_{epoch}'
+  monitor: val/loss
+  save_last: True
+  save_top_k: 0
+  every_n_epochs: 1
+
+progress_bar:
+  _target_: pytorch_lightning.callbacks.TQDMProgressBar
+  refresh_rate: ${computer.progress_bar_refresh_rate}
+
+aux_data: []
+max_epochs: 100
+data_dir: ${root_dir}/datasets
+root_dir:  ${hydra:runtime.cwd}
+experiment_name: ${dataset.name}__${model.name}
+mode: train # change that to eval to do the testing
+num_classes: 0
+areas: ['country', 'region', 'sub-region', 'city']
+class_name: null
+streetclip: False
+blur: False
+text_tuning: False
+
+hydra:
+  run:
+    dir: outputs/${hydra.job.name}/${now:%Y-%m-%d_%H-%M-%S}/${experiment_name}
+  job:
+    chdir: true

configs/dataset/baselines/im2gps.yaml

@@ -0,0 +1,16 @@
+dataset:
+  name: im2gps
+  global_batch_size: 512
+  test_dataset:
+    _partial_: true
+    _target_: data.data.Baseline
+    path: ${data_dir}/baselines/im2gps
+    which: 'im2gps'
+    transforms: ${dataset.test_transform}
+datamodule:
+  _target_: data.datamodule.BaselineDataModule
+  test_dataset: ${dataset.test_dataset}
+  global_batch_size: ${dataset.global_batch_size}
+  num_workers: ${computer.num_workers}
+  num_nodes: ${computer.num_nodes}
+  num_devices: ${computer.devices}

configs/dataset/baselines/im2gps3k.yaml

@@ -0,0 +1,16 @@
+dataset:
+  name: im2gps3k
+  global_batch_size: 512
+  test_dataset:
+    _partial_: true
+    _target_: data.data.Baseline
+    path: ${data_dir}/baselines/im2gps3k
+    which: 'im2gps3k'
+    transforms: ${dataset.test_transform}
+datamodule:
+  _target_: data.datamodule.BaselineDataModule
+  test_dataset: ${dataset.test_dataset}
+  global_batch_size: ${dataset.global_batch_size}
+  num_workers: ${computer.num_workers}
+  num_nodes: ${computer.num_nodes}
+  num_devices: ${computer.devices}

configs/dataset/baselines/yfcc4k.yaml

@@ -0,0 +1,16 @@
+dataset:
+  name: yfcc4k
+  global_batch_size: 512
+  test_dataset:
+    _partial_: true
+    _target_: data.data.Baseline
+    path: ${data_dir}/baselines/yfcc4k
+    which: 'yfcc4k'
+    transforms: ${dataset.test_transform}
+datamodule:
+  _target_: data.datamodule.BaselineDataModule
+  test_dataset: ${dataset.test_dataset}
+  global_batch_size: ${dataset.global_batch_size}
+  num_workers: ${computer.num_workers}
+  num_nodes: ${computer.num_nodes}
+  num_devices: ${computer.devices}

configs/dataset/osv5m.yaml

@@ -0,0 +1,46 @@
+defaults:
+  - train_transform: fast_clip
+  - test_transform: fast_clip
+  - _self_
+
+name: osv5m
+global_batch_size: 256
+
+train_dataset:
+  _partial_: true
+  _target_: data.data.osv5m
+  path: ${data_dir}/osv5m/
+  split: train
+  class_name: ${class_name}
+  transforms: ${dataset.train_transform}
+  aux_data: ${aux_data}
+  is_baseline: ${is_baseline}
+  areas: ${areas}
+  streetclip: ${streetclip}
+  blur: ${blur}
+
+val_dataset:
+  _partial_: true
+  _target_: data.data.osv5m
+  path: ${data_dir}/osv5m/
+  split: val
+  class_name: ${class_name}
+  transforms: ${dataset.test_transform}
+  aux_data: ${aux_data}
+  is_baseline: ${is_baseline}
+  areas: ${areas}
+  streetclip: ${streetclip}
+  blur: ${blur}
+
+test_dataset:
+  _partial_: true
+  _target_: data.data.osv5m
+  path: ${data_dir}/osv5m/
+  split: test
+  class_name: ${class_name}
+  transforms: ${dataset.test_transform}
+  aux_data: ${aux_data}
+  is_baseline: ${is_baseline}
+  areas: ${areas}
+  streetclip: ${streetclip}
+  blur: ${blur}

configs/dataset/osv5m_contrastive.yaml

@@ -0,0 +1,34 @@
+defaults:
+  - train_transform: fast_clip
+  - test_transform: fast_clip
+  - _self_
+
+name: osv5m
+global_batch_size: 256
+
+train_dataset:
+  _partial_: true
+  _target_: data.data.Contrastiveosv5m
+  path: ${data_dir}/osv5m/
+  split: train
+  class_name: ${class_name}
+  transforms: ${dataset.train_transform}
+  blur: ${blur}
+
+val_dataset:
+  _partial_: true
+  _target_: data.data.Contrastiveosv5m
+  path: ${data_dir}/osv5m/
+  split: val
+  class_name: ${class_name}
+  transforms: ${dataset.test_transform}
+  blur: ${blur}
+
+test_dataset:
+  _partial_: true
+  _target_: data.data.Contrastiveosv5m
+  path: ${data_dir}/osv5m/
+  split: test
+  class_name: ${class_name}
+  transforms: ${dataset.test_transform}
+  blur: ${blur}

configs/dataset/osv5m_contrastive_best.yaml

@@ -0,0 +1,37 @@
+defaults:
+  - train_transform: fast_clip
+  - test_transform: fast_clip
+  - _self_
+
+name: osv5m
+global_batch_size: 256
+
+train_dataset:
+  _partial_: true
+  _target_: data.data.Contrastiveosv5m
+  path: ${data_dir}/osv5m/
+  split: train
+  class_name: ${class_name}
+  transforms: ${dataset.train_transform}
+  class_name2: 'unique_region'
+  blur: ${blur}
+
+val_dataset:
+  _partial_: true
+  _target_: data.data.Contrastiveosv5m
+  path: ${data_dir}/osv5m/
+  split: val
+  class_name: ${class_name}
+  transforms: ${dataset.test_transform}
+  class_name2: 'unique_region'
+  blur: ${blur}
+
+test_dataset:
+  _partial_: true
+  _target_: data.data.Contrastiveosv5m
+  path: ${data_dir}/osv5m/
+  split: test
+  class_name: ${class_name}
+  transforms: ${dataset.test_transform}
+  class_name2: 'unique_region'
+  blur: ${blur}

configs/dataset/osv5m_text_contrastive.yaml

@@ -0,0 +1,34 @@
+defaults:
+  - train_transform: fast_clip
+  - test_transform: fast_clip
+  - _self_
+
+name: osv5m
+global_batch_size: 256
+
+train_dataset:
+  _partial_: true
+  _target_: data.data.TextContrastiveosv5m
+  path: ${data_dir}/osv5m/
+  split: train
+  class_name: ${class_name}
+  transforms: ${dataset.train_transform}
+  blur: ${blur}
+
+val_dataset:
+  _partial_: true
+  _target_: data.data.TextContrastiveosv5m
+  path: ${data_dir}/osv5m/
+  split: val
+  class_name: ${class_name}
+  transforms: ${dataset.test_transform}
+  blur: ${blur}
+
+test_dataset:
+  _partial_: true
+  _target_: data.data.TextContrastiveosv5m
+  path: ${data_dir}/osv5m/
+  split: test
+  class_name: ${class_name}
+  transforms: ${dataset.test_transform}
+  blur: ${blur}

configs/dataset/test_transform/center_crop.yaml

@@ -0,0 +1,12 @@
+_target_: torchvision.transforms.Compose
+transforms:
+  - _target_: torchvision.transforms.ToTensor
+  - _target_: utils.image_processing.CenterCrop
+    ratio: "1:1"
+  - _target_: torchvision.transforms.Resize
+    size: ${dataset.img_resolution}
+    interpolation: 3
+    antialias: true
+  - _target_: torchvision.transforms.Normalize
+    mean: 0.5
+    std: 0.5

configs/dataset/test_transform/clip.yaml

@@ -0,0 +1,2 @@
+_target_: data.transforms.ClipTransform
+split: val

configs/dataset/test_transform/fast_clip.yaml

@@ -0,0 +1,12 @@
+_target_: torchvision.transforms.Compose
+transforms:
+  - _target_: torchvision.transforms.Resize
+    size: 224
+    interpolation: 3
+    antialias: true
+  - _target_: torchvision.transforms.CenterCrop
+    size: 224
+  - _target_: torchvision.transforms.ToTensor
+  - _target_: torchvision.transforms.Normalize
+    mean: [0.48145466, 0.4578275, 0.40821073]
+    std: [0.26862954, 0.26130258, 0.27577711]

configs/dataset/test_transform/fast_resnet.yaml

@@ -0,0 +1,12 @@
+_target_: torchvision.transforms.Compose
+transforms:
+  - _target_: torchvision.transforms.Resize
+    size: 224
+    interpolation: 3
+    antialias: true
+  - _target_: torchvision.transforms.CenterCrop
+    size: 224
+  - _target_: torchvision.transforms.ToTensor
+  - _target_: torchvision.transforms.Normalize
+    mean: [0.485 ,0.456 ,0.406]
+    std: [0.229, 0.224, 0.225]

configs/dataset/test_transform/none.yaml

@@ -0,0 +1,6 @@
+_target_: torchvision.transforms.Compose
+transforms:
+  - _target_: torchvision.transforms.ToTensor
+  - _target_: torchvision.transforms.Normalize
+    mean: 0.5
+    std: 0.5

configs/dataset/train_transform/augmentation.yaml

@@ -0,0 +1,85 @@
+_target_: data.augmentation.ImageAugmentation
+names: "standard_augmentation,geometric_augmentation,clip_transform"
+
+# always apply clip_transform at the end
+clip_transform:
+  _target_: torchvision.transforms.Compose
+  transforms:
+    - _target_: torchvision.transforms.Resize
+      size: 224
+      interpolation: 3
+      antialias: true
+    - _target_: torchvision.transforms.CenterCrop
+      size: 224
+    - _target_: torchvision.transforms.ToTensor
+    - _target_: torchvision.transforms.Normalize
+      mean: [0.48145466, 0.4578275, 0.40821073]
+      std: [0.26862954, 0.26130258, 0.27577711]
+
+standard_augmentation:
+  _target_: data.augmentation.StandardAugmentation
+  # by default, we all augmentation methods
+  names: "brightness,contrast,sharpness,color,blur,gaussian_noise"
+
+  # random PIL brigtness
+  brightness:
+    _target_: data.augmentation.PillowBrightness
+    p: 0.2
+    factor_interval: [0.5, 1.5]
+
+  # random PIL contrast
+  contrast:
+    _target_: data.augmentation.PillowContrast
+    p: 0.2
+    factor_interval: [0.3, 3]
+
+  # random PIL sharpness
+  sharpness:
+    _target_: data.augmentation.PillowSharpness
+    p: 0.2
+    factor_interval: [0.5, 30.0]
+
+  # random PIL color
+  color:
+    _target_: data.augmentation.PillowColor
+    p: 0.2
+    factor_interval: [0.0, 2.0]
+
+  # random PIL blur
+  blur:
+    _target_: data.augmentation.PillowBlur
+    p: 0.2
+    factor_interval: [1, 2]
+
+  # random numpy gaussian noise
+  gaussian_noise:
+    _target_: data.augmentation.NumpyGaussianNoise
+    p: 0.2
+    factor_interval: [0.1, 0.04]
+
+geometric_augmentation:
+  _target_: data.augmentation.GeometricAugmentation
+  # by default, we all augmentation methods
+  names: "random_rotation,random_resized_crop,random_horizontal_flip"
+
+  # random rotation
+  random_rotation:
+    _target_: torchvision.transforms.RandomRotation
+    degrees: [-15, 15]
+
+  # random crop
+  random_resized_crop:
+    _target_: torchvision.transforms.RandomResizedCrop
+    scale: [0.5, 1.0]
+    ratio: [0.9, 1.1]
+    size: 224
+
+  # random horizontal flip
+  random_horizontal_flip:
+    _target_: torchvision.transforms.RandomHorizontalFlip
+    p: 0.5
+
+  # random vertical flip
+  random_vertical_flip:
+    _target_: torchvision.transforms.RandomVerticalFlip
+    p: 0.5

configs/dataset/train_transform/center_crop.yaml

@@ -0,0 +1,14 @@
+_target_: torchvision.transforms.Compose
+transforms:
+  - _target_: torchvision.transforms.ToTensor
+  - _target_: utils.image_processing.CenterCrop
+    ratio: "1:1"
+  - _target_: torchvision.transforms.Resize
+    size: ${dataset.img_resolution}
+    interpolation: 3
+    antialias: true
+  - _target_: torchvision.transforms.RandomHorizontalFlip
+    p: 0.5
+  - _target_: torchvision.transforms.Normalize
+    mean: 0.5
+    std: 0.5

configs/dataset/train_transform/clip.yaml

@@ -0,0 +1,2 @@
+_target_: data.transforms.ClipTransform
+split: val

configs/dataset/train_transform/fast_clip.yaml

@@ -0,0 +1,12 @@
+_target_: torchvision.transforms.Compose
+transforms:
+  - _target_: torchvision.transforms.Resize
+    size: 224
+    interpolation: 3
+    antialias: true
+  - _target_: torchvision.transforms.CenterCrop
+    size: 224
+  - _target_: torchvision.transforms.ToTensor
+  - _target_: torchvision.transforms.Normalize
+    mean: [0.48145466, 0.4578275, 0.40821073]
+    std: [0.26862954, 0.26130258, 0.27577711]

configs/dataset/train_transform/fast_resnet.yaml

@@ -0,0 +1,12 @@
+_target_: torchvision.transforms.Compose
+transforms:
+  - _target_: torchvision.transforms.Resize
+    size: 224
+    interpolation: 3
+    antialias: true
+  - _target_: torchvision.transforms.CenterCrop
+    size: 224
+  - _target_: torchvision.transforms.ToTensor
+  - _target_: torchvision.transforms.Normalize
+    mean: [0.485 ,0.456 ,0.406]
+    std: [0.229, 0.224, 0.225]

configs/dataset/train_transform/none.yaml

@@ -0,0 +1,7 @@
+_target_: torchvision.transforms.Compose
+transforms:
+  - _target_: torchvision.transforms.Resize
+    size: 224
+    interpolation: 3
+    antialias: true
+  - _target_: torchvision.transforms.ToTensor

configs/exp/DinoV2.yaml

@@ -0,0 +1,18 @@
+# @package _global_
+
+defaults:
+  - override /model: regression
+  - override /model/network/backbone: dinov2_vitl14
+  - _self_
+
+model:
+  optimizer:
+    optim:
+      lr: 0.0002
+      weight_decay: 0.0001
+
+is_baseline: false
+max_epochs: 30
+
+dataset:
+  global_batch_size: 2048

configs/exp/ResNet.yaml

@@ -0,0 +1,21 @@
+# @package _global_
+
+defaults:
+  - override /model: regression
+  - override /dataset/test_transform: fast_resnet
+  - override /dataset/train_transform: fast_resnet 
+  - override /model.network.mid: mlp_resnet
+  - override /model/network/backbone: ResNet50
+  - _self_
+
+model:
+  optimizer:
+    optim:
+      lr: 0.0002
+      weight_decay: 0.0001
+
+is_baseline: false
+max_epochs: 30
+
+dataset:
+  global_batch_size: 2048

configs/exp/base_model.yaml

@@ -0,0 +1,19 @@
+# @package _global_
+
+defaults:
+  - override /model: regression
+  - override /model/network/backbone: openclip_B_32
+  - _self_
+
+model:
+  name: base_model
+  optimizer:
+    optim:
+      lr: 0.0002
+      weight_decay: 0.0001
+
+is_baseline: false
+max_epochs: 30
+
+dataset:
+  global_batch_size: 2048

configs/exp/best_model.yaml

@@ -0,0 +1,25 @@
+# @package _global_
+
+defaults:
+  - override /dataset: osv5m_contrastive_best
+  - override /model: hybrid
+  - override /model/network: best_backbone
+  - override /model/network/backbone: clip_L_14_DataComp
+  - override /model/network/mid: mlp_hybrid
+  - override /model/loss: best_model
+  - _self_
+
+class_name: 'quadtree_10_1000'
+is_baseline: false
+max_epochs: 30
+
+model:
+  name: best_model
+  optimizer:
+    optim:
+      lr: 2e-4
+      weight_decay: 0.0001
+    backbone_lr: 2e-5
+
+dataset:
+  global_batch_size: 2048

configs/exp/classification_area.yaml

@@ -0,0 +1,19 @@
+# @package _global_
+
+defaults:
+  - override /model: classification
+  - override /model/network/backbone: openclip_B_32
+  - _self_
+
+class_name: 'area'
+model:
+  optimizer:
+    optim:
+      lr: 0.0002
+      weight_decay: 0.0001
+
+is_baseline: false
+max_epochs: 15
+
+dataset:
+  global_batch_size: 2048

configs/exp/classification_cell.yaml

@@ -0,0 +1,19 @@
+# @package _global_
+
+defaults:
+  - override /model: classification
+  - override /model/network/backbone: openclip_B_32
+  - _self_
+
+class_name: quadtree_10_1000
+model:
+  optimizer:
+    optim:
+      lr: 0.0002
+      weight_decay: 0.0001
+
+is_baseline: false
+max_epochs: 15
+
+dataset:
+  global_batch_size: 2048

configs/exp/classification_cell_hier.yaml

@@ -0,0 +1,20 @@
+# @package _global_
+
+defaults:
+  - override /model: classification
+  - override /model/network/backbone: openclip_B_32
+  - override /model/loss: cls_hier_quad
+  - _self_
+
+class_name: quadtree_10_1000
+model:
+  optimizer:
+    optim:
+      lr: 0.0002
+      weight_decay: 0.0001
+
+is_baseline: false
+max_epochs: 15
+
+dataset:
+  global_batch_size: 2048

configs/exp/classification_city.yaml

@@ -0,0 +1,19 @@
+# @package _global_
+
+defaults:
+  - override /model: classification
+  - override /model/network/backbone: openclip_B_32
+  - _self_
+
+class_name: 'city'
+model:
+  optimizer:
+    optim:
+      lr: 0.0002
+      weight_decay: 0.0001
+
+is_baseline: false
+max_epochs: 15
+
+dataset:
+  global_batch_size: 2048

configs/exp/classification_city_hier.yaml

@@ -0,0 +1,20 @@
+# @package _global_
+
+defaults:
+  - override /model: classification
+  - override /model/network/backbone: openclip_B_32
+  - override /model/loss: cls_hier
+  - _self_
+
+class_name: 'city'
+model:
+  optimizer:
+    optim:
+      lr: 0.0002
+      weight_decay: 0.0001
+
+is_baseline: false
+max_epochs: 15
+
+dataset:
+  global_batch_size: 2048

configs/exp/classification_country.yaml

@@ -0,0 +1,19 @@
+# @package _global_
+
+defaults:
+  - override /model: classification
+  - override /model/network/backbone: openclip_B_32
+  - _self_
+
+class_name: 'country'
+model:
+  optimizer:
+    optim:
+      lr: 0.0002
+      weight_decay: 0.0001
+
+is_baseline: false
+max_epochs: 15
+
+dataset:
+  global_batch_size: 2048

configs/exp/classification_region copy.yaml

@@ -0,0 +1,19 @@
+# @package _global_
+
+defaults:
+  - override /model: classification
+  - override /model/network/backbone: openclip_B_32
+  - _self_
+
+class_name: 'region'
+model:
+  optimizer:
+    optim:
+      lr: 0.0002
+      weight_decay: 0.0001
+
+is_baseline: false
+max_epochs: 15
+
+dataset:
+  global_batch_size: 2048

configs/exp/classification_region.yaml

@@ -0,0 +1,19 @@
+# @package _global_
+
+defaults:
+  - override /model: classification
+  - override /model/network/backbone: openclip_B_32
+  - _self_
+
+class_name: 'region'
+model:
+  optimizer:
+    optim:
+      lr: 0.0002
+      weight_decay: 0.0001
+
+is_baseline: false
+max_epochs: 15
+
+dataset:
+  global_batch_size: 2048

configs/exp/clip_L_14_DataComp.yaml

@@ -0,0 +1,18 @@
+# @package _global_
+
+defaults:
+  - override /model: regression
+  - override /model/network/backbone: clip_L_14_DataComp
+  - _self_
+
+model:
+  optimizer:
+    optim:
+      lr: 0.0002
+      weight_decay: 0.0001
+
+is_baseline: false
+max_epochs: 30
+
+dataset:
+  global_batch_size: 2048

configs/exp/clip_L_14_Laion.yaml

@@ -0,0 +1,18 @@
+# @package _global_
+
+defaults:
+  - override /model: regression
+  - override /model/network/backbone: openclip_L_14
+  - _self_
+
+model:
+  optimizer:
+    optim:
+      lr: 0.0002
+      weight_decay: 0.0001
+
+is_baseline: false
+max_epochs: 30
+
+dataset:
+  global_batch_size: 2048

configs/exp/clip_L_14_OpenAI.yaml

@@ -0,0 +1,18 @@
+# @package _global_
+
+defaults:
+  - override /model: regression
+  - override /model/network/backbone: clip_L_14
+  - _self_
+
+model:
+  optimizer:
+    optim:
+      lr: 0.0002
+      weight_decay: 0.0001
+
+is_baseline: false
+max_epochs: 30
+
+dataset:
+  global_batch_size: 2048

configs/exp/clip_bigG_14_Laion.yaml

@@ -0,0 +1,18 @@
+# @package _global_
+
+defaults:
+  - override /model: regression
+  - override /model/network/backbone: openclip_bigG_14
+  - _self_
+
+model:
+  optimizer:
+    optim:
+      lr: 0.0002
+      weight_decay: 0.0001
+
+is_baseline: false
+max_epochs: 30
+
+dataset:
+  global_batch_size: 2048

configs/exp/contrastive_area.yaml

@@ -0,0 +1,20 @@
+# @package _global_
+
+defaults:
+  - override /dataset: osv5m_contrastive
+  - override /model: regression
+  - override /model/network: contrastive_unfrozen_backbone
+  - override /model/network/backbone: openclip_B_32
+  - override /model/loss: contrastive
+  - _self_
+
+model:
+  optimizer:
+    optim:
+      lr: 2e-4
+      weight_decay: 0.0001
+    backbone_lr: 2e-5
+
+class_name: area
+is_baseline: false
+max_epochs: 30

configs/exp/contrastive_cell.yaml

@@ -0,0 +1,20 @@
+# @package _global_
+
+defaults:
+  - override /dataset: osv5m_contrastive
+  - override /model: regression
+  - override /model/network: contrastive_unfrozen_backbone
+  - override /model/network/backbone: openclip_B_32
+  - override /model/loss: contrastive
+  - _self_
+
+model:
+  optimizer:
+    optim:
+      lr: 2e-4
+      weight_decay: 0.0001
+    backbone_lr: 2e-5
+
+class_name: quadtree_10_1000
+is_baseline: false
+max_epochs: 30

configs/exp/contrastive_city.yaml

@@ -0,0 +1,20 @@
+# @package _global_
+
+defaults:
+  - override /dataset: osv5m_contrastive
+  - override /model: regression
+  - override /model/network: contrastive_unfrozen_backbone
+  - override /model/network/backbone: openclip_B_32
+  - override /model/loss: contrastive
+  - _self_
+
+model:
+  optimizer:
+    optim:
+      lr: 2e-4
+      weight_decay: 0.0001
+    backbone_lr: 2e-5
+
+class_name: city
+is_baseline: false
+max_epochs: 30

configs/exp/contrastive_country.yaml

@@ -0,0 +1,20 @@
+# @package _global_
+
+defaults:
+  - override /dataset: osv5m_contrastive
+  - override /model: regression
+  - override /model/network: contrastive_unfrozen_backbone
+  - override /model/network/backbone: openclip_B_32
+  - override /model/loss: contrastive
+  - _self_
+
+model:
+  optimizer:
+    optim:
+      lr: 2e-4
+      weight_decay: 0.0001
+    backbone_lr: 2e-5
+
+class_name: country
+is_baseline: false
+max_epochs: 30

configs/exp/contrastive_region.yaml

@@ -0,0 +1,20 @@
+# @package _global_
+
+defaults:
+  - override /dataset: osv5m_contrastive
+  - override /model: regression
+  - override /model/network: contrastive_unfrozen_backbone
+  - override /model/network/backbone: openclip_B_32
+  - override /model/loss: contrastive
+  - _self_
+
+model:
+  optimizer:
+    optim:
+      lr: 2e-4
+      weight_decay: 0.0001
+    backbone_lr: 2e-5
+
+class_name: region
+is_baseline: false
+max_epochs: 30

configs/exp/contrastive_text.yaml

@@ -0,0 +1,22 @@
+# @package _global_
+
+defaults:
+  - override /dataset: osv5m_text_contrastive
+  - override /model: text_tuning
+  - override /model/network/backbone: openclip_B_32
+  - _self_
+
+model:
+  network:
+    backbone:
+      instance:
+        _target_: models.networks.backbones.CLIPText
+  optimizer:
+    optim:
+      lr: 0.0002
+      weight_decay: 0.0001
+
+is_baseline: false
+class_name: city
+text_tuning: True
+max_epochs: 30

configs/exp/eval_best_model.yaml

@@ -0,0 +1,29 @@
+# @package _global_
+
+defaults:
+  - override /dataset: osv5m_contrastive_best
+  - override /model: hybrid
+  - override /model/network: best_backbone
+  - override /model/network/backbone: clip_L_14_DataComp
+  - override /model/network/mid: mlp_hybrid
+  - _self_
+
+class_name: 'quadtree_10_1000'
+is_baseline: false
+max_epochs: 30
+mode: 'eval'
+
+model:
+  name: best_model
+  optimizer:
+    optim:
+      lr: 2e-4
+      weight_decay: 0.0001
+    backbone_lr: 2e-5
+  network:
+    head:
+      instance:
+        quadtree_path: ${root_dir}/quadtree_10_1000.csv
+
+dataset:
+  global_batch_size: 2048