image test

app.py

@@ -3,23 +3,24 @@ import pandas as pd
 from annoy import AnnoyIndex
 from deepface import DeepFace
 import gdown
+from similarity_finder.get_similar_images import get_similar_images
 
-def getSimilarImages():
-    return []
 
-def my_inference_function(name):
-  return "Hello " + name + "!"
+def my_inference_function(image_path):
+  #global df
+  #get_similar_images(df, embedding, age, gender, race)
+  return "Hello " + image_path + "!"
 
-gradio_interface = gradio.Interface(
+gradio_interface = gradio.Interface(  
   fn = my_inference_function,
-  inputs = "text",
+  inputs = gr.Image(type="filepath"),
   outputs = "text"
 )
 
 print("downloading dataset")
-id = "1z6amrk21M-4N39OZLOsi4WWjAdBVSjh7"
-gdown.download(id=id, output="final_dataframe.parquet.gzip", quiet=False)
+#id = "1z6amrk21M-4N39OZLOsi4WWjAdBVSjh7"
+#gdown.download(id=id, output="final_dataframe.parquet.gzip", quiet=False)
 
 print("loading dataframe")
-df = pd.read_parquet("final_dataframe.parquet.gzip")
-gradio_interface.launch()
+#df = pd.read_parquet("final_dataframe.parquet.gzip")
+gradio_interface.launch(share=True)

similarity_finder/get_similar_images.py

@@ -0,0 +1,100 @@
+import pandas as pd
+import numpy as np
+from annoy import AnnoyIndex
+from deepface import DeepFace
+import argparse
+
+FILE_PATHS = {
+    "dataframe": {
+        "id": "1HFxHX2RkEr7_yVHnA-qk5Lj8CxOWrUda",
+        "name": "final_embeddings_clusters.parquet.gzip",
+        "path": "preprocessed_files"
+    },
+    "AnnoyIndex_Saved_File": {
+        "id": "14uIgsVAiGolTy3-TGWrUUXqEzJqh3ZMl",
+        "name": "CACD2000_refined_images_embeddings_clusters.ann",
+            "path": "preprocessed_files"
+        }
+    }
+
+def download_file(file_id, file_name, save_path):
+    """Function to generate the urls for given params"""
+    url = r"""wget --load-cookies /tmp/cookies.txt "https://docs.google.com/uc?export=download&confirm=$(wget --quiet --save-cookies /tmp/cookies.txt --keep-session-cookies --no-check-certificate 'https://docs.google.com/uc?export=download&id={FILE_ID}' -O- | sed -rn 's/.*confirm=([0-9A-Za-z_]+).*/\1\n/p')&id={FILE_ID}" -O {SAVE_PATH}/{FILE_NAME} && rm -rf /tmp/cookies.txt""".format(
+        FILE_ID=file_id, FILE_NAME=file_name, SAVE_PATH=save_path
+    )
+    os.system(url)
+
+
+"""
+Prepare env for using this file, use this function
+if running code without backend.
+"""
+def download_required_files():
+    programs = []
+    for key, details in MODEL_PATHS.items():
+        if not os.path.exists(details["path"]):
+            os.makedirs(details["path"])
+            proc = Process(target=download_file, args=(
+                    details["id"], details["name"], details["path"],))
+            programs.append(proc)
+            proc.start()
+
+    for proc in programs:
+        proc.join()
+
+    return "Environent Ready!"
+
+def get_similar_images_annoy(t, df, img_index, n=1000, max_dist=1.0):
+    vid, face  = df.iloc[img_index, [0, 1]]
+    # print(vid)
+    similar_img_ids, dist = t.get_nns_by_item(img_index, n+1, include_distances=True)
+    similar_img_ids = [s for s,d in zip(similar_img_ids, dist) if (d <= max_dist and df['race'][s]==df['race'][img_index] and df['gender'][s]==df['gender'][img_index] and int(df['age'][s])<int(df['age'][img_index])+5 and int(df['age'][s])>int(df['age'][img_index])-5)][1:]  # first item is always its own video
+    return vid, vid, df.iloc[similar_img_ids], dist
+
+def get_sample_n_similar(t, df, sample_idx):
+    output_images = []
+    vid, face, similar, distances = get_similar_images_annoy(t, df, sample_idx)
+    list_plot = [face] + similar['face'].values.tolist()
+    list_cluster = [df.iloc[sample_idx]['cluster']] + similar['cluster'].values.tolist()
+    for face, cluster, dist in zip(list_plot, list_cluster, distances):
+      try:
+        output_images.append(f'{face.split("/")[-1][:-4]}.jpg')
+      except:
+        continue
+    return output_images[1:10]
+
+def add_to_dataframe(embedding, age, gender, race, dataframe):
+    embedding_json = {}
+    embedding_json['face'] = "user_image"
+    #embedding_objs = DeepFace.represent(img_path = image_path)
+    embedding_json['embedding'] = embedding
+    embedding_json['age'] = age
+    embedding_json['gender'] = gender
+    embedding_json['race'] = race
+    _ = pd.json_normalize(embedding_json)
+    dataframe = pd.concat([_, dataframe], sort=False, ignore_index=True)
+    
+    return dataframe
+
+def get_similar_images(df, embedding, age, gender, race):
+    df = add_to_dataframe(embedding, age, gender, race, df)
+
+    f = len(df['embedding'][0])
+    t = AnnoyIndex(f, metric='euclidean')
+    ntree = 50
+    
+    for i, vector in enumerate(df['embedding']):
+        t.add_item(i, vector)
+       
+    _  = t.build(ntree)
+    
+    results = get_sample_n_similar(t, df, 0)
+    return t, results
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser()
+    parser.parse_args()
+    parser.add_argument("image_file_path", help="Enter the apth of the image file that you need similar images for")
+    args = parser.parse_args()
+    image_path = str(args.image_file_path)
+    get_similar_images(image_path)

similarity_finder/requirements.txt

@@ -0,0 +1,3 @@
+annoy
+pandas
+numpy

similarity_finder/shell.nix

@@ -0,0 +1,19 @@
+with import <nixpkgs> { };
+
+let pythonPackages = python310Packages;
+in pkgs.mkShell rec {
+	name = "similarity-finder-env";
+	venvDir = "./env";
+	buildInputs = [
+		stdenv.cc.cc.lib
+        stdenv.cc
+		pythonPackages.python
+		pythonPackages.venvShellHook
+	];
+
+	LD_LIBRARY_PATH = pkgs.lib.makeLibraryPath buildInputs;
+
+	postVenvCreation = ''
+		pip install -r requirements.txt
+	'';
+}