import streamlit as st
import degirum as dg
from PIL import Image
import torch
import numpy as np
import torch.nn.functional as F
import clip
import cv2

# Compute the cosine similarity between two vectors.
def cosine_similarity(a, b):
    return np.dot(a, b) / (np.linalg.norm(a) * np.linalg.norm(b))

def compute_text_embeddings(text_prompts):
    device = "cuda" if torch.cuda.is_available() else "cpu"
    clip_model, _ = clip.load("RN50", device=device)
    
    text_embeddings = []
    for text_prompt in text_prompts:
        text = clip.tokenize(text_prompt).to(device)
        text_embedding = clip_model.encode_text(text)
        text_embeddings.append(text_embedding.cpu().detach().numpy().tolist())
    
    return text_embeddings

zoo=dg.connect(dg.CLOUD,zoo_url='https://cs.degirum.com/degirum/kvk_upload_test', token="dg_ZPNjDPnAPpoZiykpFz2CnoYaAYf6ewBxwDCBo")#,token=st.secrets["DG_TOKEN"])

st.title('DeGirum CLIP model Demo')

with st.sidebar:
    st.header('Specify Model Options Below')
    prompts = st.text_area("Enter text prompts (comma-separated):", value="People Running, People Talking, People Fighting, People Laughing, People Dancing")
    prompts = [prompt.strip() for prompt in prompts.split(',')]
st.text('Upload an image. Then click on the submit button')
with st.form("model_form"):
    uploaded_file=st.file_uploader('input image')
    submitted = st.form_submit_button("Submit")
    embeddings = compute_text_embeddings(prompts)
    if submitted:
        model=zoo.load_model('clip--224x224_float_openvino_cpu_4',
                             input_image_format = "RAW"                                    
                            )
        image = Image.open(uploaded_file)
        opencv_image = np.array(image)
        opencv_image = cv2.cvtColor(opencv_image, cv2.COLOR_RGB2BGR)
        predictions=model(opencv_image).results[0]["data"]
        dg_cloud_output_reshaped = predictions.reshape(-1)
        similarities = [cosine_similarity(dg_cloud_output_reshaped, np.array(embedding).reshape(-1)) for embedding in embeddings]
        similarities_tensor = torch.tensor(similarities, dtype=torch.float32)
        softmax_scores = F.softmax(similarities_tensor, dim=0)   
        max_index = torch.argmax(softmax_scores).item()
        st.image(image, caption="Uploaded Image", use_column_width=True)
        for index, prompt in enumerate(prompts):
            st.write(f"{prompt} - {softmax_scores[index]*100:.2f}%")

        # st.write(predictions.results)