Hugging Face's logo

Spaces:
MatanKriel
/
Food_Recommender
Sleeping

App Files Community
MatanKriel commited on
Commit
3e810e8
·
verified ·
1 Parent(s): 19567a7

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +11 -17
README.md CHANGED
@@ -59,24 +59,16 @@ Based on the plots above, **we deleted "bad" images** that were:
59
  * Too Bright/Washed out (Avg Pixel Intensity > 245)
60
  * Extreme Aspect Ratios (Too stretched or squashed, AR > 3.0)
61
 
62
- ### 3. Advanced Feature Engineering
63
- After removing the garbage data, we engineered deeper visual features to assess image content:
64
-
65
- * **Sharpness Score:** Used Laplacian Variance to find blurry photos.
66
- * **Dominant Color (Hue):** Analyzed color clusters (e.g., Green for Salads vs. Red for Pizza).
67
- * **Texture Complexity:** Calculated pixel standard deviation to distinguish smooth vs. complex foods.
68
-
69
-
70
- ![image](https://cdn-uploads.huggingface.co/production/uploads/67dfcd96d01eab4618a66f78/0QMOkOCATUfePwu_-nm0z.png)
71
 
72
  ---
73
 
74
- ## ⚔️ Part 2: Model Comparison (CLIP vs. SigLIP)
75
- To ensure the best search results, we ran a "Challenger" test between two leading multimodal models.
76
 
77
  ### The Contestants:
78
  1. **Baseline:** OpenAI CLIP (`clip-vit-base-patch32`)
79
  2. **Challenger:** Google SigLIP (`siglip-base-patch16-224`)
 
80
 
81
  ### The Evaluation:
82
  We compared them using **Silhouette Scores** (measuring how distinct the food clusters are) and a visual "Taste Test" (checking nearest neighbors for specific dishes).
@@ -88,12 +80,13 @@ We compared them using **Silhouette Scores** (measuring how distinct the food cl
88
  We queried both models with the same image to see which returned more accurate similar foods.
89
 
90
 
91
- ![image](https://cdn-uploads.huggingface.co/production/uploads/67dfcd96d01eab4618a66f78/R4biFno1FUizlVRLRVCqM.png)
 
92
 
93
  ---
94
 
95
  ## 🧠 Part 3: Embeddings & Clustering
96
- Using the winning model (**SigLIP**), we generated 768-dimensional vectors for the entire dataset. We applied dimensionality reduction to visualize how the AI groups food concepts.
97
 
98
  * **Algorithm:** K-Means Clustering (k=101 categories).
99
  * **Visualization:**
@@ -101,10 +94,8 @@ Using the winning model (**SigLIP**), we generated 768-dimensional vectors for t
101
  * **t-SNE:** To see local groupings (e.g., "Sushi" clusters separately from "Burgers").
102
 
103
 
104
- ![image](https://cdn-uploads.huggingface.co/production/uploads/67dfcd96d01eab4618a66f78/MT92BvvwToLxk83X0Yd12.png)
105
-
106
 
107
- ![image](https://cdn-uploads.huggingface.co/production/uploads/67dfcd96d01eab4618a66f78/KyMPVz6VUsIGq2IMEZRYl.png)
108
 
109
  ---
110
 
@@ -114,10 +105,13 @@ The final product is a **Gradio** web application hosted on Hugging Face Spaces.
114
  1. **Image-to-Image:** Upload a photo (e.g., a burger) -> The app embeds it using SigLIP -> Finds the nearest 3 visual matches.
115
  2. **Text-to-Image:** Type "Spicy Tacos" -> The app finds images matching that description.
116
 
 
 
 
117
  ### How to Run Locally
118
  1. **Clone the repository:**
119
  ```bash
120
- git clone [https://huggingface.co/spaces/YOUR_USERNAME/Food-Match](https://huggingface.co/spaces/YOUR_USERNAME/Food-Match)
121
  cd Food-Match
122
  ```
123
  2. **Install dependencies:**
 
59
  * Too Bright/Washed out (Avg Pixel Intensity > 245)
60
  * Extreme Aspect Ratios (Too stretched or squashed, AR > 3.0)
61
 
 
 
 
 
 
 
 
 
 
62
 
63
  ---
64
 
65
+ ## ⚔️ Part 2: Model Comparison (CLIP vs. SigLIP vs metaclip)
66
+ To ensure the best search results, we ran a "Challenger" test between three leading multimodal models.
67
 
68
  ### The Contestants:
69
  1. **Baseline:** OpenAI CLIP (`clip-vit-base-patch32`)
70
  2. **Challenger:** Google SigLIP (`siglip-base-patch16-224`)
71
+ 3. **Challenger:** Facebook MetaCLIP": ("facebook/metaclip-b32-400m)
72
 
73
  ### The Evaluation:
74
  We compared them using **Silhouette Scores** (measuring how distinct the food clusters are) and a visual "Taste Test" (checking nearest neighbors for specific dishes).
 
80
  We queried both models with the same image to see which returned more accurate similar foods.
81
 
82
 
83
+
84
+ ![image](https://cdn-uploads.huggingface.co/production/uploads/67dfcd96d01eab4618a66f78/Yz1YyU-eGcH9806Kg6PiF.png)
85
 
86
  ---
87
 
88
  ## 🧠 Part 3: Embeddings & Clustering
89
+ Using the winning model (**SigLIP**), We applied dimensionality reduction to visualize how the AI groups food concepts.
90
 
91
  * **Algorithm:** K-Means Clustering (k=101 categories).
92
  * **Visualization:**
 
94
  * **t-SNE:** To see local groupings (e.g., "Sushi" clusters separately from "Burgers").
95
 
96
 
 
 
97
 
98
+ ![image](https://cdn-uploads.huggingface.co/production/uploads/67dfcd96d01eab4618a66f78/i3qZepniP0HqGQ8m5H-7K.png)
99
 
100
  ---
101
 
 
105
  1. **Image-to-Image:** Upload a photo (e.g., a burger) -> The app embeds it using SigLIP -> Finds the nearest 3 visual matches.
106
  2. **Text-to-Image:** Type "Spicy Tacos" -> The app finds images matching that description.
107
 
108
+ ## Note
109
+ The application is running the clip model even though the sigLip model won, sigLip was to big to be run on the hugging face space free tier
110
+
111
  ### How to Run Locally
112
  1. **Clone the repository:**
113
  ```bash
114
+ git clone [https://huggingface.co/spaces/YOUR_USERNAME/Food_Recommender](https://huggingface.co/spaces/YOUR_USERNAME/Food-Match)
115
  cd Food-Match
116
  ```
117
  2. **Install dependencies:**