Improve readme

README.md

@@ -35,7 +35,7 @@ The model can be used as a **binary (true/false) classifier if desired, or you c
 
 ### Global Performance
 
-| Category | Freepik | Falconasai | Adamcodd |
+| Category | Freepik | Falconsai | Adamcodd |
 |----------|-------------|------------------|----------------|
 | High     | 99.54%     | 97.92%          | 98.62%         |
 | Medium   | 97.02%     | 78.54%          | 91.65%         |
@@ -43,10 +43,23 @@ The model can be used as a **binary (true/false) classifier if desired, or you c
 | Neutral  | 99.87%     | 99.27%          | 98.37%         |
 
 
-**Notes:**
-* Values represent the percentage of correct predictions. For Falconasai and AdamCodd, these are True/False predictions, while for the Freepik model, they represent detecting at least 'low' level content.
-* AdamCodd and Falconasai show lower accuracy on intermediate categories because they are designed for binary classification only.
-* Our model outperforms both models in accuracy and granularity.
+In the table below, the results are obtained as follows:
+ 
+* For the **Falconsai and AdamCodd** models:
+    * A prediction is considered correct if the image is labeled "low", "medium", or "high" and the model returns true.
+    * If the label is "neutral", the correct output should be false.
+
+* For the **Freepik model**:
+    * If the image label is "low", "medium", or "high", the model should return at least "low".
+    * If the label is "neutral", the correct output should be "neutral".
+
+
+**Conclusions:**
+
+* Our model **outperforms AdamCodd and Falconsai in accuracy**. It is entirely fair to compare them on the "high" and "neutral" labels.
+* Our model **offers greater granularity**. It is not only suitable for detecting "high" and "neutral" content, but also performs excellently at identifying "low" and "medium" NSFW content.
+    * Falconsai may classify some "medium" and "low" images as not NSFW but mark others as safe for work(SFW), which could lead to unexpected results.
+    * AdamCodd classifies both "low" and "medium" categories as NSFW, which may not be desirable depending on your use case. Furthermore, a 10% of images in low and medium are considered SFW.
 
 ### Accuracy by AI Content
 
@@ -56,25 +69,18 @@ The following tables show detection accuracy percentages across different NSFW c
 
 #### AI-Generated Content
 
-| Category | Freepik Model | Falconasai Model | Adamcodd Model |
+| Category | Freepik Model | Falconsai Model | Adamcodd Model |
 |----------|-------------|------------------|----------------|
 | High     | 100.00%     | 84.00%          | 92.00%         |
 | Medium   | 96.15%      | 69.23%          | 96.00%         |
 | Low      | 100.00%     | 35.71%          | 92.86%         |
 | Neutral  | 100.00%     | 100.00%         | 66.67%         |
 
-### Non-AI-Generated Content
-
-| Category | Freepik Model | Falconasai Model | Adamcodd Model |
-|----------|-------------|------------------|----------------|
-| High     | 100.00%     | 100.00%         | 100.00%        |
-| Medium   | 100.00%     | 70.00%          | 100.00%        |
-| Low      | 100.00%     | 0.00%           | 100.00%        |
-| Neutral  | 100.00%     | 88.88%          | 100.00%        |
 
-**Notes:**
+**Conclusions:**
+* **Avoid using Falconsai for AI-generated content** to prevent prediction errors.
+* **Our model is the best option to detect NSFW content in AI-generated content**.  
 
-* Due to the small dataset size, some categories show 100.00% accuracy in some cases, expect less accuracy in real world
 
 ## Usage
 
@@ -95,7 +101,7 @@ detector = NSFWDetector(dtype=torch.bfloat16, device="cuda")
 # Load and classify an image
 image = Image.open("your_image")
 
-# Check if the image contains NSFW content sentivity level medium
+# Check if the image contains NSFW content sentivity level medium or higher
 is_nsfw = detector.is_nsfw(image, "medium")
 
 # Get probability scores for all categories
@@ -261,7 +267,7 @@ Example output:
 
 **Notes:**
 * The model has been trained in bf16 so it is **recommended to use it in bf16**.
-* **Using torch tensor**: The speed using torch tensor is not achieved using pipeline. Avoid it in production.
+* **Using torch tensor**: The speed using torch tensor is not achieved using pipeline. Avoid pipeline use in production.
 * Measurements taken on **NVIDIA RTX 3090**, expect better metrics in more powerful servers.
 * Throughput increases with larger batch sizes due to better GPU utilization. Consider your use case when selecting batch size.
 * Optimizations listed are suggestions that could further improve performance.