application bones

app.py

@@ -1,8 +1,43 @@
 import streamlit as st
-from transformers import pipeline
 from PIL import Image
+import sys
+import zipfile
+from fastai import *
+import numpy as np
+import pandas as pd 
+import os
+import timm
+from timm import create_model 
+from fastai.vision.all import *
 
-pipeline = pipeline(task="image-classification", use_auth_token=True, model="blanar/pawpularity")
+torch.device('cpu')
+def seed_everything(seed):
+    random.seed(seed)
+    os.environ['PYTHONHASHSEED'] = str(seed)
+    np.random.seed(seed)
+    torch.manual_seed(seed)
+    torch.backends.cudnn.deterministic = True
+seed_everything(42)
+
+test = {
+    'Id': ['2022-12-21 00.49.46.jpg', '2022-12-21 00.49.46.jpg'],
+    'Eyes': [0.3, 0.2],
+}
+train_df = pd.DataFrame(test)
+print(train_df)
+dls = DataBlock(blocks=(ImageBlock, CategoryBlock),
+                get_x=ColReader('Id'),
+                get_y=ColReader('Eyes'),
+                splitter=RandomSplitter(0.2),
+                item_tfms=Resize(224),
+                batch_tfms=setup_aug_tfms([Brightness(), Contrast(), Hue(), Saturation(), Flip(size=224)]),
+               )
+
+paw_dls = dls.dataloaders(train_df, batch_size=8, seed=12, device='cpu')
+test = paw_dls.test_dl(train_df)
+learn = cnn_learner(paw_dls, models.resnet50, pretrained=False, metrics=error_rate) 
+learn.to('cpu')
+catanddog = learn.load('../dog')
 
 st.title("Hot Dog? Or Not?")
 
@@ -13,8 +48,22 @@ if file_name is not None:
 
     image = Image.open(file_name)
     col1.image(image, use_column_width=True)
-    predictions = pipeline(image)
-
+    pred = catanddog.predict(image)[1]
+    print(pred)
+    def metric_rmse(input,target):
+        return 100*torch.sqrt(F.mse_loss(F.sigmoid(input.flatten()), target))
+    model = create_model('swin_large_patch4_window7_224', pretrained=False, num_classes=1)
+    learn = Learner(paw_dls, model, loss_func = BCEWithLogitsLossFlat(),  metrics=metric_rmse)
+    # learner = learner.load(f'/kaggle/input/puredog/pure-dog-{fold}')
+    if pred == 0:
+        learn.load('../archive-3/pure-dog-1')
+        pred = learn.predict(image)
+        print("Score for this doggy:", int(pred[2] * 100))
+    if pred == 1:
+        for i in range(10):
+            learn.load(f'../archive-2/pure-cat-{i}')
+            pred = learn.predict(image)
+            print(pred)
+        print("Score for this catto:", min(int(pred[2] * 100 + 35), 100))
     col2.header("Probabilities")
-    for p in predictions:
-        col2.subheader(f"{ p['label'] }: { round(p['score'] * 100, 1)}%")
+    col2.subheader(f"{ pred[2] }: { round(pred[2] * 100, 1)}%")

dog.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:48e47295238ba0b456d20e57b4f971c8c77d395443fa97b4f7e958c2f476be80
+size 120168877

pure-cat-0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a3a0451aaffb232bd79dbb8f5b33ee204afec651e3aa3aca4c18ea18419b25ac
+size 781699177

pure-dog-5.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e2c47c357d0f2c26576969f412229f9f8b9ddd304fe8ed7c971d8efe4c82a9d0
+size 781699177

requirements.txt

@@ -1,2 +1,3 @@
 transformers
-torch
+torch
+fastai

test.py

@@ -0,0 +1,55 @@
+import sys
+import zipfile
+from fastai import *
+import numpy as np
+import pandas as pd 
+import os
+import timm
+from timm import create_model 
+from fastai.vision.all import *
+
+torch.device('cpu')
+def seed_everything(seed):
+    random.seed(seed)
+    os.environ['PYTHONHASHSEED'] = str(seed)
+    np.random.seed(seed)
+    torch.manual_seed(seed)
+    torch.backends.cudnn.deterministic = True
+seed_everything(42)
+
+test = {
+    'Id': ['2022-12-21 00.49.46.jpg', '2022-12-21 00.49.46.jpg'],
+    'Eyes': [0.3, 0.2],
+}
+train_df = pd.DataFrame(test)
+print(train_df)
+dls = DataBlock(blocks=(ImageBlock, CategoryBlock),
+                get_x=ColReader('Id'),
+                get_y=ColReader('Eyes'),
+                splitter=RandomSplitter(0.2),
+                item_tfms=Resize(224),
+                batch_tfms=setup_aug_tfms([Brightness(), Contrast(), Hue(), Saturation(), Flip(size=224)]),
+               )
+paw_dls = dls.dataloaders(train_df, batch_size=8, seed=12, device='cpu')
+test = paw_dls.test_dl(train_df)
+learn = cnn_learner(paw_dls, models.resnet50, pretrained=False, metrics=error_rate) 
+learn.to('cpu')
+catanddog = learn.load('../dog')
+image = 'photo-1529778873920-4da4926a72c2.jpeg'
+pred = catanddog.predict(image)[1]
+print(pred)
+def metric_rmse(input,target):
+    return 100*torch.sqrt(F.mse_loss(F.sigmoid(input.flatten()), target))
+model = create_model('swin_large_patch4_window7_224', pretrained=False, num_classes=1)
+learn = Learner(paw_dls, model, loss_func = BCEWithLogitsLossFlat(),  metrics=metric_rmse)
+# learner = learner.load(f'/kaggle/input/puredog/pure-dog-{fold}')
+if pred == 0:
+    learn.load('../archive-3/pure-dog-1')
+    pred = learn.predict(image)
+    print("Score for this doggy:", int(pred[2] * 100))
+if pred == 1:
+    for i in range(10):
+        learn.load(f'../archive-2/pure-cat-{i}')
+        pred = learn.predict(image)
+        print(pred)
+    print("Score for this catto:", min(int(pred[2] * 100 + 35), 100))