Update readme
Browse files- readme_.md +14 -6
readme_.md
CHANGED
|
@@ -2,13 +2,13 @@
|
|
| 2 |
|
| 3 |
This app compares 3 image classification approaches on car images:
|
| 4 |
|
| 5 |
-
- Fine-tuned ViT model (
|
| 6 |
- Zero-shot CLIP (`openai/clip-vit-large-patch14`)
|
| 7 |
- OpenAI vision model (GPT-4o image classification)
|
| 8 |
|
| 9 |
## Dataset Used For Training
|
| 10 |
|
| 11 |
-
- Hugging Face dataset:
|
| 12 |
- The Stanford Cars dataset contains 196 fine-grained classes (car make/model/year combinations). We group them into 9 brand-level classes for a cleaner, more visually meaningful classification task.
|
| 13 |
- Number of classes: `9`
|
| 14 |
- Classes: `BMW`, `Dodge`, `Ferrari`, `Ford`, `Jeep`, `Lamborghini`, `Porsche`, `Rolls-Royce`, `Toyota`
|
|
@@ -24,7 +24,7 @@ This app compares 3 image classification approaches on car images:
|
|
| 24 |
|
| 25 |
## Trained Model
|
| 26 |
|
| 27 |
-
- Hugging Face model link:
|
| 28 |
- Base model: `google/vit-base-patch16-224`
|
| 29 |
- Only the final classification head was fine-tuned (all other layers frozen).
|
| 30 |
- Trainable parameters: ~4,614 out of ~85.8M total.
|
|
@@ -41,7 +41,7 @@ This app compares 3 image classification approaches on car images:
|
|
| 41 |
|
| 42 |
## Hugging Face Space
|
| 43 |
|
| 44 |
-
- App link:
|
| 45 |
|
| 46 |
## Example Image Results
|
| 47 |
|
|
@@ -51,5 +51,13 @@ The table below reports the true class and Top-3 predictions for ViT, CLIP, and
|
|
| 51 |
|---|---|---|---|---|
|
| 52 |
| `Dodge.jpg` | `Dodge` | BMW (0.3564), Dodge (0.2218), Rolls-Royce (0.1807) | Dodge (0.9432), Jeep (0.0393), Lamborghini (0.0078) | Dodge (1.00) |
|
| 53 |
| `Ferrari.jpg` | `Ferrari` | Ferrari (0.6007), Lamborghini (0.2946), Ford (0.0296) | Ferrari (0.9958), Lamborghini (0.0032), Ford (0.0004) | Ferrari (1.00) |
|
| 54 |
-
| `BMW.jpg` | `BMW` | BMW (0.2737), Porsche (0.1800), Dodge (0.1630) | BMW (0.9969), Porsche (0.0014), Ferrari (0.0007) | BMW (
|
| 55 |
-
| `Porsche.jpg` | `Porsche` | BMW (0.5858), Dodge (0.2040), Toyota (0.0667) | Porsche (0.9887), Lamborghini (0.0047), Dodge (0.0022) | Porsche (1.00) |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2 |
|
| 3 |
This app compares 3 image classification approaches on car images:
|
| 4 |
|
| 5 |
+
- Fine-tuned ViT model (`nenzilea/car-classification`)
|
| 6 |
- Zero-shot CLIP (`openai/clip-vit-large-patch14`)
|
| 7 |
- OpenAI vision model (GPT-4o image classification)
|
| 8 |
|
| 9 |
## Dataset Used For Training
|
| 10 |
|
| 11 |
+
- Hugging Face dataset: https://huggingface.co/datasets/tanganke/stanford_cars
|
| 12 |
- The Stanford Cars dataset contains 196 fine-grained classes (car make/model/year combinations). We group them into 9 brand-level classes for a cleaner, more visually meaningful classification task.
|
| 13 |
- Number of classes: `9`
|
| 14 |
- Classes: `BMW`, `Dodge`, `Ferrari`, `Ford`, `Jeep`, `Lamborghini`, `Porsche`, `Rolls-Royce`, `Toyota`
|
|
|
|
| 24 |
|
| 25 |
## Trained Model
|
| 26 |
|
| 27 |
+
- Hugging Face model link: https://huggingface.co/nenzilea/car-classification
|
| 28 |
- Base model: `google/vit-base-patch16-224`
|
| 29 |
- Only the final classification head was fine-tuned (all other layers frozen).
|
| 30 |
- Trainable parameters: ~4,614 out of ~85.8M total.
|
|
|
|
| 41 |
|
| 42 |
## Hugging Face Space
|
| 43 |
|
| 44 |
+
- App link: https://huggingface.co/spaces/nenzilea/car-classification
|
| 45 |
|
| 46 |
## Example Image Results
|
| 47 |
|
|
|
|
| 51 |
|---|---|---|---|---|
|
| 52 |
| `Dodge.jpg` | `Dodge` | BMW (0.3564), Dodge (0.2218), Rolls-Royce (0.1807) | Dodge (0.9432), Jeep (0.0393), Lamborghini (0.0078) | Dodge (1.00) |
|
| 53 |
| `Ferrari.jpg` | `Ferrari` | Ferrari (0.6007), Lamborghini (0.2946), Ford (0.0296) | Ferrari (0.9958), Lamborghini (0.0032), Ford (0.0004) | Ferrari (1.00) |
|
| 54 |
+
| `BMW.jpg` | `BMW` | BMW (0.2737), Porsche (0.1800), Dodge (0.1630) | BMW (0.9969), Porsche (0.0014), Ferrari (0.0007) | BMW (0.95), Porsche (0.001), Dodge (0.001), Ferrari (0.001), Ford (0.001) |
|
| 55 |
+
| `Porsche.jpg` | `Porsche` | BMW (0.5858), Dodge (0.2040), Toyota (0.0667) | Porsche (0.9887), Lamborghini (0.0047), Dodge (0.0022) | Porsche (1.00) |
|
| 56 |
+
|
| 57 |
+
## Model Comparison Summary
|
| 58 |
+
|
| 59 |
+
| Model | Approach | Strengths | Weaknesses |
|
| 60 |
+
|---|---|---|---|
|
| 61 |
+
| **Custom ViT** | Supervised fine-tuning on 9 car brands | High accuracy on known brands | Only classifies the 9 trained brands |
|
| 62 |
+
| **CLIP** | Zero-shot with brand name as text prompt | No training needed, flexible labels | Lower accuracy; may confuse visually similar brands |
|
| 63 |
+
| **OpenAI GPT-4o** | LLM vision with natural language prompt | Strong reasoning, handles unusual angles | API cost, latency, black-box |
|