PerceptCLIP/PerceptCLIP_Emotions

PreceptCLIP-Emotions is a model designed to predict the emotions that an image evokes in users. This is the official model from the paper "Don't Judge Before You CLIP: A Unified Approach for Perceptual Tasks". We apply LoRA adaptation on the CLIP visual encoder with an additional MLP head. Our model achieves state-of-the-art results.

Training Details

• Dataset: EmoSet
• Architecture: CLIP Vision Encoder (ViT-L/14) with LoRA adaptation
• Loss Function: Cross Entropy Loss
• Optimizer: AdamW
• Learning Rate: 0.0001
• Batch Size: 32

Requirements

• python=3.9.15
• cudatoolkit=11.7
• torchvision=0.14.0
• transformers=4.45.2
• peft=0.14.0

Usage

To use the model for inference:

from torchvision import transforms
import torch
from PIL import Image
from huggingface_hub import hf_hub_download

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

# Load model
model_path = hf_hub_download(repo_id="PerceptCLIP/PerceptCLIP_Emotions", filename="perceptCLIP_Emotions.pth")
model = torch.load(model_path).to(device).eval()

# Emotion label mapping
idx2label = {
    0: "amusement",
    1: "awe",
    2: "contentment",
    3: "excitement",
    4: "anger",
    5: "disgust",
    6: "fear",
    7: "sadness"
}

# Preprocessing function
def emo_preprocess():
    transform = transforms.Compose([
        transforms.Resize(224),
        transforms.CenterCrop(size=(224, 224)),
        transforms.ToTensor(),
        transforms.Normalize(mean=(0.48145466, 0.4578275, 0.40821073), std=(0.26862954, 0.26130258, 0.27577711)),
    ])
    return transform

# Load an image
image = Image.open("image_path.jpg").convert("RGB")
image = emo_preprocess()(image).unsqueeze(0).to(device)

# Run inference
with torch.no_grad():
    outputs = model(image)
    _, predicted = outputs.max(1)  # Get the class index

# Get emotion label
predicted_emotion = idx2label[predicted.item()]
print(f"Predicted Emotion: {predicted_emotion}")