Add model weights as submodule

.gitmodules

@@ -0,0 +1,3 @@
+[submodule "galaxy-zoo-generation"]
+	path = galaxy-zoo-generation
+	url = https://huggingface.co/vitaliykinakh/galaxy-zoo-generation

galaxy-zoo-generation

@@ -0,0 +1 @@
+Subproject commit af017a826b2be3dec5f364ac4e232ede6cc0e04f

src/app/compare_models.py

@@ -1,4 +1,3 @@
-from pathlib import Path
 import math
 
 import streamlit as st
@@ -14,7 +13,7 @@ from src.models import ConditionalGenerator as InfoSCC_GAN
 from src.models.big.BigGAN2 import Generator as BigGAN2Generator
 from src.models import ConditionalDecoder as cVAE
 from src.data import get_labels_train, make_galaxy_labels_hierarchical
-from src.utils import download_file, sample_labels
+from src.utils import sample_labels
 
 
 device = params.device
@@ -76,25 +75,14 @@ def load_model(model_type: str):
                         y_size=params.shape_label,
                         z_size=params.noise_dim)
 
-        if not Path(params.path_infoscc_gan).exists():
-            download_file(params.drive_id_infoscc_gan, params.path_infoscc_gan)
-
         ckpt = torch.load(params.path_infoscc_gan, map_location=torch.device('cpu'))
         g.load_state_dict(ckpt['g_ema'])
     elif model_type == 'BigGAN':
         g = BigGAN2Generator()
-
-        if not Path(params.path_biggan).exists():
-            download_file(params.drive_id_biggan, params.path_biggan)
-
         ckpt = torch.load(params.path_biggan, map_location=torch.device('cpu'))
         g.load_state_dict(ckpt)
     elif model_type == 'cVAE':
         g = cVAE()
-
-        if not Path(params.path_cvae).exists():
-            download_file(params.drive_id_cvae, params.path_cvae)
-
         ckpt = torch.load(params.path_cvae, map_location=torch.device('cpu'))
         g.load_state_dict(ckpt)
     else:
@@ -107,9 +95,6 @@ def load_model(model_type: str):
 def get_labels() -> torch.Tensor:
     path_labels = params.path_labels
 
-    if not Path(path_labels).exists():
-        download_file(params.drive_id_labels, path_labels)
-
     labels_train = get_labels_train(path_labels)
     return labels_train
 

src/app/explore_biggan.py

@@ -1,5 +1,4 @@
 import math
-from pathlib import Path
 
 import streamlit as st
 import numpy as np
@@ -12,7 +11,7 @@ from src.app.questions import q1, q1_options, q2, q2_options, q3, q3_options, q4
     q6, q6_options, q7, q7_options, q8, q8_options, q9, q9_options, q10, q10_options, q11, q11_options
 from src.models.big.BigGAN2 import Generator as BigGAN2Generator
 from src.data import get_labels_train, make_galaxy_labels_hierarchical
-from src.utils import download_file, sample_labels
+from src.utils import sample_labels
 
 
 # global parameters
@@ -25,7 +24,6 @@ dim_z = params.dim_z
 bs = 16  # number of samples to generate
 n_cols = int(math.sqrt(bs))
 model_path = params.path_biggan
-drive_id = params.drive_id_biggan
 path_labels = params.path_labels
 
 # manual labels
@@ -90,9 +88,6 @@ def get_eps(n: int) -> torch.Tensor:
 
 @st.cache
 def get_labels() -> torch.Tensor:
-    if not Path(path_labels).exists():
-        download_file(params.drive_id_labels, path_labels)
-
     labels_train = get_labels_train(path_labels)
     return labels_train
 
@@ -102,10 +97,6 @@ def app():
 
     st.title('Explore BigGAN')
     st.markdown('This demo shows BigGAN for conditional galaxy generation')
-
-    if not Path(model_path).exists():
-        download_file(drive_id, model_path)
-
     model = load_model(model_path)
     eps = get_eps(bs)
     labels_train = get_labels()

src/app/explore_cvae.py

@@ -1,5 +1,4 @@
 import math
-from pathlib import Path
 
 import streamlit as st
 import numpy as np
@@ -12,7 +11,7 @@ from src.app.questions import q1, q1_options, q2, q2_options, q3, q3_options, q4
     q6, q6_options, q7, q7_options, q8, q8_options, q9, q9_options, q10, q10_options, q11, q11_options
 from src.models import ConditionalDecoder
 from src.data import get_labels_train, make_galaxy_labels_hierarchical
-from src.utils import download_file, sample_labels
+from src.utils import sample_labels
 
 
 # global parameters
@@ -25,7 +24,6 @@ dim_z = params.dim_z
 bs = 16  # number of samples to generate
 n_cols = int(math.sqrt(bs))
 model_path = params.path_cvae
-drive_id = params.drive_id_cvae
 path_labels = params.path_labels
 
 # manual labels
@@ -90,9 +88,6 @@ def get_eps(n: int) -> torch.Tensor:
 
 @st.cache
 def get_labels() -> torch.Tensor:
-    if not Path(path_labels).exists():
-        download_file(params.drive_id_labels, path_labels)
-
     labels_train = get_labels_train(path_labels)
     return labels_train
 
@@ -103,9 +98,6 @@ def app():
     st.title('Explore cVAE')
     st.markdown('This demo shows cVAE for conditional galaxy generation')
 
-    if not Path(model_path).exists():
-        download_file(drive_id, model_path)
-
     model = load_model(model_path)
     eps = get_eps(bs)
     labels_train = get_labels()

src/app/explore_infoscc_gan.py

@@ -1,4 +1,3 @@
-from pathlib import Path
 import math
 
 import numpy as np
@@ -12,7 +11,7 @@ from src.app.questions import q1, q1_options, q2, q2_options, q3, q3_options, q4
     q6, q6_options, q7, q7_options, q8, q8_options, q9, q9_options, q10, q10_options, q11, q11_options
 from src.models import ConditionalGenerator
 from src.data import get_labels_train, make_galaxy_labels_hierarchical
-from src.utils import download_file, sample_labels
+from src.utils import sample_labels
 
 # global parameters
 device = params.device
@@ -27,7 +26,6 @@ y_type = params.y_type
 bs = 16  # number of samples to generate
 n_cols = int(math.sqrt(bs))
 model_path = params.path_infoscc_gan  # path to the model
-drive_id = params.drive_id_infoscc_gan   # google drive id of the model
 path_labels = params.path_labels
 
 # manual labels
@@ -87,8 +85,6 @@ def load_model(model_path: str) -> ConditionalGenerator:
 
 @st.cache
 def get_labels() -> torch.Tensor:
-    if not Path(path_labels).exists():
-        download_file(params.drive_id_labels, path_labels)
     labels_train = get_labels_train(path_labels)
     return labels_train
 
@@ -100,9 +96,6 @@ def app():
     st.markdown('This demo shows InfoSCC-GAN for conditional galaxy generation')
     st.subheader(r'<- Use sidebar to explore $z_1, ..., z_k$ latent variables')
 
-    if not Path(model_path).exists():
-        download_file(drive_id, model_path)
-
     model = load_model(model_path)
     eps = model.sample_eps(bs).to(device)
     labels_train = get_labels()

src/app/interpolate_labels.py

@@ -1,4 +1,3 @@
-from pathlib import Path
 import math
 
 import numpy as np
@@ -12,7 +11,7 @@ from src.models import ConditionalGenerator as InfoSCC_GAN
 from src.models.big.BigGAN2 import Generator as BigGAN2Generator
 from src.models import ConditionalDecoder as cVAE
 from src.data import get_labels_train
-from src.utils import download_file, sample_labels
+from src.utils import sample_labels
 
 
 device = params.device
@@ -31,26 +30,14 @@ def load_model(model_type: str):
         g = InfoSCC_GAN(size=params.size,
                         y_size=params.shape_label,
                         z_size=params.noise_dim)
-
-        if not Path(params.path_infoscc_gan).exists():
-            download_file(params.drive_id_infoscc_gan, params.path_infoscc_gan)
-
         ckpt = torch.load(params.path_infoscc_gan, map_location=torch.device('cpu'))
         g.load_state_dict(ckpt['g_ema'])
     elif model_type == 'BigGAN':
         g = BigGAN2Generator()
-
-        if not Path(params.path_biggan).exists():
-            download_file(params.drive_id_biggan, params.path_biggan)
-
         ckpt = torch.load(params.path_biggan, map_location=torch.device('cpu'))
         g.load_state_dict(ckpt)
     elif model_type == 'cVAE':
         g = cVAE()
-
-        if not Path(params.path_cvae).exists():
-            download_file(params.drive_id_cvae, params.path_cvae)
-
         ckpt = torch.load(params.path_cvae, map_location=torch.device('cpu'))
         g.load_state_dict(ckpt)
     else:
@@ -62,10 +49,6 @@ def load_model(model_type: str):
 @st.cache
 def get_labels() -> torch.Tensor:
     path_labels = params.path_labels
-
-    if not Path(path_labels).exists():
-        download_file(params.drive_id_labels, path_labels)
-
     labels_train = get_labels_train(path_labels)
     return labels_train
 

src/app/params.py

@@ -12,14 +12,7 @@ n_basis = 6   # size of additional z vectors in InfoSCC-GAN
 y_type = 'real'  # type of labels in InfoSCC-GAN
 dim_z = 128  # z vector size in BigGAN and cVAE
 
-path_infoscc_gan = './models/InfoSCC-GAN/generator.pt'
-drive_id_infoscc_gan = '1_kIujc497OH0ZJ7PNPwS5_otNlS7jMLI'
-
-path_biggan = './models/BigGAN/generator.pth'
-drive_id_biggan = '1sMSDdnQ5GjHcno5knHTDSKAKhhoHh_4z'
-
-path_cvae = './models/CVAE/generator.pth'
-drive_id_cvae = '17FmLvhwXq8PQMrD1CtjqyoAy5BobYMTE'
-
-path_labels = './data/training_solutions_rev1.csv'
-drive_id_labels = '1dzsB_HdGtmSHE4pCppamISpFaJBfPF7E'
+path_infoscc_gan = './galaxy-zoo-generation/models/InfoSCC-GAN/generator.pt'
+path_biggan = './galaxy-zoo-generation/models/BigGAN/generator.pth'
+path_cvae = './galaxy-zoo-generation/models/CVAE/generator.pth'
+path_labels = './galaxy-zoo-generation/data/training_solutions_rev1.csv'