DVC added and working webapp with inference

Aptfile

@@ -0,0 +1 @@
+https://github.com/iterative/dvc/releases/download/2.8.3/dvc_2.8.3_amd64.deb

DVC-heroku-deployment.md

@@ -0,0 +1,21 @@
+We need to give Heroku the ability to pull in data from DVC upon app start up. We will install a [buildpack](https://elements.heroku.com/buildpacks/heroku/heroku-buildpack-apt) that allows the installation of apt-files and then define the Aptfile that contains a path to DVC. I.e., in the CLI run:
+
+```
+heroku buildpacks:add --index 1 heroku-community/apt
+```
+
+Then in your root project folder create a file called `Aptfile` that specifies the release of DVC you want installed, https://github.com/iterative/dvc/releases/download/2.8.3/dvc_2.8.3_amd64.deb
+
+Add the following code block to your **streamlit_app.py**:
+
+```python
+import os
+
+if "DYNO" in os.environ and os.path.isdir(".dvc"):
+    os.system("dvc config core.no_scm true")
+    if os.system(f"dvc pull {model} {image}") != 0:
+        exit("dvc pull failed")
+    os.system("rm -r .dvc .apt/usr/lib/dvc")
+```
+
+Reference: [Heroku ML](https://github.com/GuilhermeBrejeiro/deploy_ML_model_Heroku_FastAPI)

app/app_savta.py

@@ -1,9 +1,61 @@
 import numpy as np
 import torch
 import sys
+import os
 from fastai.vision.all import *
 import gradio as gr
 
+############## DVC ################################
+
+PROD_MODEL_PATH = "src/models"
+TRAIN_PATH = "src/data/processed/train/bathroom"
+TEST_PATH = ""src/data/processed/test/bathroom""
+
+if "DYNO" in os.environ and os.path.isdir(".dvc"):
+    print("Running DVC")
+    os.system("dvc config cache.type copy")
+    os.system("dvc config core.no_scm true")
+    if os.system(f"dvc pull {PROD_MODEL_PATH} {TRAIN_PATH } {TEST_PATH }") != 0:
+        exit("dvc pull failed")
+    os.system("rm -r .dvc .apt/usr/lib/dvc")
+
+
+############## Inference ##############################
+
+class ImageImageDataLoaders(DataLoaders):
+    """Basic wrapper around several `DataLoader`s with factory methods for Image to Image problems"""
+    @classmethod
+    @delegates(DataLoaders.from_dblock)
+    def from_label_func(cls, path, filenames, label_func, valid_pct=0.2, seed=None, item_transforms=None,
+                        batch_transforms=None, **kwargs):
+        """Create from list of `fnames` in `path`s with `label_func`."""
+        datablock = DataBlock(blocks=(ImageBlock(cls=PILImage), ImageBlock(cls=PILImageBW)),
+                              get_y=label_func,
+                              splitter=RandomSplitter(valid_pct, seed=seed),
+                              item_tfms=item_transforms,
+                              batch_tfms=batch_transforms)
+        res = cls.from_dblock(datablock, filenames, path=path, **kwargs)
+        return res
+
+
+def get_y_fn(x):
+    y = str(x.absolute()).replace('.jpg', '_depth.png')
+    y = Path(y)
+
+    return y
+
+
+def create_data(data_path):
+    fnames = get_files(data_path/'train', extensions='.jpg')
+    data = ImageImageDataLoaders.from_label_func(data_path/'train', seed=42, bs=4, num_workers=0, filenames=fnames, label_func=get_y_fn)
+    return data
+
+data = create_data(Path('src/data/processed'))
+learner = unet_learner(data,resnet34, metrics=rmse, wd=1e-2, n_out=3, loss_func=MSELossFlat(), path='src/')
+learner.load('model')
+
+################### Gradio Web APP ################################
+
 title = "SavtaDepth WebApp"
 description = "Savta Depth is a collaborative Open Source Data Science project for monocular depth estimation - Turn 2d photos into 3d photos. To test the model and code please check out the link bellow."
 article = "<p style='text-align: center'><a href='https://dagshub.com/OperationSavta/SavtaDepth' target='_blank'>SavtaDepth Project from OperationSavta</a></p><p style='text-align: center'><a href='https://colab.research.google.com/drive/1XU4DgQ217_hUMU1dllppeQNw3pTRlHy1?usp=sharing' target='_blank'>Google Colab Demo</a></p></center></p>"
@@ -14,22 +66,16 @@ examples = [
 ]
 favicon = "examples/favicon.ico"
 thumbnail = "examples/SavtaDepth.png"
-learner = unet_learner(resnet34, path='src/')
-learner.load('model')
-def sepia(input_img):
-    sepia_filter = np.array(
-        [[0.393, 0.769, 0.189], [0.349, 0.686, 0.168], [0.272, 0.534, 0.131]]
-    )
-    sepia_img = input_img.dot(sepia_filter.T)
-    sepia_img /= sepia_img.max()
-    return sepia_img
 
+def sepia(input_img):
+   
+    return PILImageBW.create((learner.predict(input_img))[0]).convert('L')
 
 
 def main():
-	iface = gr.Interface(sepia, gr.inputs.Image(shape=(640,480)), "image", title = title, description = description, article = article, examples = examples,theme ="peach",thumbnail=thumbnail)
+	iface = gr.Interface(sepia, gr.inputs.Image(shape=(640,480),type='numpy'), "image", title = title, description = description, article = article, examples = examples,theme ="peach",thumbnail=thumbnail)
 
-	iface.launch(favicon_path=favicon,auth=("admin", "dagshubsota123"),server_name="0.0.0.0")
+	iface.launch(favicon_path=favicon,server_name="0.0.0.0",server_port=8080)
 # enable_queue=True,auth=("admin", "pass1234")
 
 if __name__ == '__main__':

requirements.txt

@@ -1,11 +1,11 @@
 -f https://download.pytorch.org/whl/cpu/torch_stable.html
 #dvc
-#fastai
+fastai
 torch==1.10.2+cpu
-h5py
+#h5py
 opencv-python-headless
 tqdm
-numpy
+#numpy
 #scikit-learn
 #dagshub
 #tables