add recommendation.py

app.py

@@ -9,6 +9,8 @@ from PIL import Image
 from numpy import random
 from pathlib import Path
 import torch.backends.cudnn as cudnn
+
+from recommendation import SimilarityRecommender
 from models.experimental import attempt_load
 
 from utils.datasets import LoadStreams, LoadImages
@@ -35,6 +37,7 @@ def print_output(txt_path):
 def detect_logo(img, model, confidence):
     if model == 'Yologo':
         model = 'best_logo'  # Naming Convention for yolov7 See output file of https://www.kaggle.com/code/owaiskhan9654/training-yolov7-on-kaggle-on-custom-dataset/data
+    recommender = SimilarityRecommender("./TopBrands.xlsx")
     parser = argparse.ArgumentParser()
     parser.add_argument('--weights', nargs='+', type=str, default=model + ".pt", help='model.pt path(s)')
     parser.add_argument('--source', type=str, default='inference/', help='source')
@@ -119,14 +122,18 @@ def detect_logo(img, model, confidence):
             txt_path = str(save_dir / 'labels' / p.stem) + ('' if dataset.mode == 'image' else f'_{frame}')  # img.txt
             s += '%gx%g ' % img.shape[2:]
             gn = torch.tensor(im0.shape)[[1, 0, 1, 0]]
+            all_classes_detected = []
+            all_recommendations = []
             if len(det):
                 det[:, :4] = scale_coords(img.shape[2:], det[:, :4], im0.shape).round()
 
                 for c in det[:, -1].unique():
                     n = (det[:, -1] == c).sum()
                     s += f"{n} {names[int(c)]}{'s' * (n > 1)}, "
-
                 for *xyxy, conf, cls in reversed(det):
+                    name_class = names[int(cls.item())]
+                    all_classes_detected.append(name_class)
+                    all_recommendations.append(recommender.make_recommendation(name_class))
                     if save_txt:
                         xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) / gn).view(-1).tolist()
                         line = (names[int(cls.item())], *xywh, conf) if opt.save_conf else (names[int(cls.item())], *xywh)
@@ -165,7 +172,7 @@ def detect_logo(img, model, confidence):
     print(f'Done. ({time.time() - t0:.3f}s)')
     img_output = Image.fromarray(im0[:, :, ::-1])
     txt_output = print_output(txt_path)
-    return img_output, txt_output
+    return img_output, txt_output, "Recommendations : "+str(set(all_recommendations))
 
 examples1 = [["exemple1.jpg", "Yologo", 0.60], ["exemple2.jpg", "Yologo", 0.60],
              ["exemple3.jpg", "Yologo", 0.60 ], ["exemple4.jpg", "Yologo", 0.60],
@@ -176,8 +183,13 @@ examples1 = [["exemple1.jpg", "Yologo", 0.60], ["exemple2.jpg", "Yologo", 0.60],
 Top_Title = "<center>Yolov7 🚀 Custom Trained by ISAI Team (Capgemini) </center> Logo detection model"
 Custom_description="Yologo can detect 44 brands : A.P.C, Abercrombie, Acne Studio, Adidas, Aigle, Airness, Armarni, Balenciaga, Bulgari, Cacharel, Calvin Klein, Carhatt, Champion, Columbia, Converse, Gap, Chanel, Lacoste, Levis, Louis vuitton, Moncler, Mossimo, New Balance, Obey, Pepe jeans, Ralph Lauren, Prada, Rolex, Seiko, Stussy, The North Face, Timberland, Tom Ford, Tommy Hilfiger, Uniqlo, Valentino, Versace, Zara, Asics, Le coq, Oakley, Nike, Puma, Reebok "
 
-gr.Interface(detect_logo, [gr.Image(type="pil"), gr.Dropdown(value="Yologo",
-                                                               choices=["Yologo"]), gr.Slider(0.10, 1.0, value=0.60)], [gr.Image(type="pil"), "text"],
+# Recommendation#
+
+
+gr.Interface(detect_logo, [gr.Image(type="pil"),
+                           gr.Dropdown(value="Yologo",choices=["Yologo"]),
+                           gr.Slider(0.10, 1.0, value=0.60)],
+                           [gr.Image(type="pil"), "text", "text"],
              title=Top_Title,
             description=Custom_description,
              examples=examples1, cache_examples=True).launch()

recommendation.py

@@ -0,0 +1,29 @@
+import numpy as np # linear algebra
+import pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)
+from sklearn.metrics.pairwise import cosine_similarity
+import re
+
+class SimilarityRecommender():
+
+    def __init__(self, file):
+        data = pd.read_excel(file, index_col=0)
+        columns_brands = [re.sub(r'[^a-zA-Z ]', '', brand.upper()) for brand in list(data.index)]
+        ex = pd.DataFrame(cosine_similarity(data), index=columns_brands,
+                          columns=columns_brands)
+        for brand in ex.index:
+            ex.loc[brand, brand] = np.nan
+        self.similarity_matrix = ex.assign(best_similarity=ex.idxmax())['best_similarity']
+
+    def make_recommendation(self, item):
+        key = re.sub(r'[^a-zA-Z ]', '', item.upper())
+        if key not in self.similarity_matrix.index:
+            print(self.similarity_matrix.index)
+            raise ValueError(f'{key} not in matrix')
+        return self.similarity_matrix[key]
+
+
+if __name__ == '__main__':
+    rec = SimilarityRecommender("./TopBrands.xlsx")
+    print(rec.make_recommendation("louis vuitton-1"))
+    print(rec.similarity_matrix)
+    print(rec.make_recommendation("Lacoste"))