ViT_based_movie_identification_by_frame_lr1e-4_3epochs

Movie Identification by Frame (ViT)

This model is designed to identify movies based on a single frame. It is a fine-tuned version of the Vision Transformer (ViT), specifically the google/vit-base-patch16-224-in21k architecture.

Model Description

The model was trained to classify frames into specific movie titles from Movie Identification Dataset [800 Movies].

• Model type: Image Classification
• Architecture: Vision Transformer (ViT)
• Base Model: google/vit-base-patch16-224-in21k

Training Hyperparameters

The following settings were used during fine-tuning:

• Learning Rate: 1e-4
• Epochs: 3
• Batch Size: 64
• Optimizer: AdamW

How to use

 from transformers import ViTImageProcessor, ViTForImageClassification
 
# 1. Downloading the processor and model from repository
model_id = "callmealina/ViT_based_movie_identification_by_frame_lr1e-4_3epochs"
processor = ViTImageProcessor.from_pretrained(model_id)
model = ViTForImageClassification.from_pretrained(model_id)

# 2. Preprocess a single frame
image = Image.open("frame_0027.jpg").convert("RGB")
inputs = processor(images=image, return_tensors="pt")

3. Inference
model.eval()
with torch.no_grad():
  outputs = model(**inputs)
  logits = outputs.logits
  predicted_class_idx = logits.argmax(-1).item()

  # 4. Print the result (name of a film)
  predicted_label = model.config.id2label[predicted_class_idx]
  print(f"Predicted Movie: {predicted_label}")