app.py

import streamlit as st
import torch
from PIL import Image

from models import IndividualLandmarkViT
from utils import VisualizeAttentionMaps
from utils.transform_utils import make_test_transforms

st.title("PdiscoFormer Part Discovery Visualizer")
# Instructions
st.write("First of all choose a model from the dropdown list."
         "If you choose to upload an image, the part assignment will be visualized. "
         "The model is trained to discover parts of the salient objects in the image depending on the training dataset.")
st.write("If you choose one of the CUB or NABirds models, please choose a bird image.")
st.write("If you choose one of the Flower models, please choose a flower image.")
st.write("If you choose one of the PartImageNet models, please choose an image of classes from PartImageNet like land animals/birds/cars/bottles/airplanes.")

model_options = ["ananthu-aniraj/pdiscoformer_cub_k_8", "ananthu-aniraj/pdiscoformer_cub_k_16",
                 "ananthu-aniraj/pdiscoformer_cub_k_4", "ananthu-aniraj/pdiscoformer_part_imagenet_ood_k_8",
                 "ananthu-aniraj/pdiscoformer_part_imagenet_ood_k_25",
                 "ananthu-aniraj/pdiscoformer_part_imagenet_ood_k_50",
                 "ananthu-aniraj/pdiscoformer_flowers_k_2", "ananthu-aniraj/pdiscoformer_flowers_k_4",
                 "ananthu-aniraj/pdiscoformer_flowers_k_8", "ananthu-aniraj/pdiscoformer_nabirds_k_4",
                 "ananthu-aniraj/pdiscoformer_nabirds_k_8", "ananthu-aniraj/pdiscoformer_nabirds_k_11",
                 "ananthu-aniraj/pdiscoformer_pimagenet_seg_k_8", "ananthu-aniraj/pdiscoformer_pimagenet_seg_k_16",
                 "ananthu-aniraj/pdiscoformer_pimagenet_seg_k_25", "ananthu-aniraj/pdiscoformer_pimagenet_seg_k_41",
                 "ananthu-aniraj/pdiscoformer_pimagenet_seg_k_50"]
model_name = st.selectbox("Select a model", model_options)
if model_name is not None:
    if "cub" in model_name or "nabirds" in model_name:
        image_size = 518
    else:
        image_size = 224
    # Set the device
    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
    # Load the model
    model = IndividualLandmarkViT.from_pretrained(model_name, input_size=image_size).eval().to(device)
    num_parts = model.num_landmarks
    amap_vis = VisualizeAttentionMaps(num_parts=num_parts + 1, bg_label=num_parts)
    test_transforms = make_test_transforms(image_size)


    image_name = st.file_uploader("Upload an image", type=["jpg", "jpeg", "png"])  # Upload an image
    if image_name is not None:
        image = Image.open(image_name).convert("RGB")
        image_tensor = test_transforms(image).unsqueeze(0).to(device)
        with torch.no_grad():
            maps, scores = model(image_tensor)

        coloured_map = amap_vis.show_maps(image_tensor, maps)
        st.image(coloured_map, caption="Attention Map", use_column_width=True)