DocExplore_DEMO / app.py
CHSTR's picture
actualizando el app
9435ab3
raw
history blame
11.9 kB
from html import escape
from io import BytesIO
import base64
from multiprocessing.dummy import Pool
from PIL import Image, ImageDraw
import streamlit as st
import pandas as pd
import numpy as np
import torch
# from transformers import CLIPProcessor, CLIPModel
# from transformers import OwlViTProcessor, OwlViTForObjectDetection
# from transformers.image_utils import ImageFeatureExtractionMixin
import pickle as pkl
# sketches
from streamlit_drawable_canvas import st_canvas
from PIL import Image, ImageOps
from torchvision import transforms
# model
import os
# No reconoce la carpeta que esta dos niveles abajo src
from src.model_LN_prompt import Model
from src.options import opts
from datasets import load_dataset
DEBUG = False
if DEBUG:
MODEL = "vit-base-patch32"
else:
MODEL = "vit-large-patch14-336"
CLIP_MODEL = f"openai/clip-{MODEL}"
OWL_MODEL = f"google/owlvit-base-patch32"
if not DEBUG and torch.cuda.is_available():
device = torch.device("cuda")
else:
device = torch.device("cpu")
HEIGHT = 350
N_RESULTS = 5
from huggingface_hub import hf_hub_download,login
token = os.getenv("HUGGINGFACE_TOKEN")
# Autentica usando el token
login(token=token)
color = st.get_option("theme.primaryColor")
if color is None:
color = (0, 255, 0)
else:
color = tuple(int(color.lstrip("#")[i: i + 2], 16) for i in (0, 2, 4))
@st.cache_resource
def load():
# Descargamos el dataset
dataset = load_dataset("CHSTR/docexplore")
print(dataset)
print(dataset['features'])
#local_dir = "./"
#dataset.save_to_disk(local_dir)
path_images = dataset['features']['image']['filename']
path_model = hf_hub_download(repo_id="CHSTR/DocExplore", filename="epoch=16-mAP=0.66_triplet.ckpt")#"models/epoch=16-mAP=0.66_triplet.ckpt"
model = Model()
model_checkpoint = torch.load(path_model, map_location=device) # 'model_60k_images_073.ckpt' -> modelo entrenado con 60k imagenes sin pidinet
model.load_state_dict(model_checkpoint['state_dict']) # 'modified_model_083.ckpt' -> modelo entrenado con 60k imagenes con pidinet
model.eval() # 'original_model_083.ckpt' -> modelo original entrenado con 60k imagenes con pidinet
print("Modelo cargado exitosamente")
embeddings_file_1 = hf_hub_download(repo_id="CHSTR/DocExplore", filename="dino_flicker_docexplore_groundingDINO.pkl")
embeddings_file_0 = hf_hub_download(repo_id="CHSTR/DocExplore", filename="docexp_embeddings.pkl")
embeddings = {
0: pkl.load(open(embeddings_file_0, "rb")),
1: pkl.load(open(embeddings_file_1, "rb"))
}
# embeddings = {
# 0: pkl.load(open("docexp_embeddings.pkl", "rb")),
# 1: pkl.load(open("dino_flicker_docexplore_groundingDINO.pkl", "rb"))
# }
# Actualizar los paths de las imágenes en los embeddings
#for i in range(len(embeddings[0])):
# print(embeddings[0][i])
#embeddings[0][i] = (embeddings[0][i][0], path_images + "/".join(embeddings[0][i][1].split("/")[:-3]))
#for i in range(len(embeddings[1])):
# print(embeddings[1][i])
#embeddings[1][i] = (embeddings[1][i][0], path_images + "/".join(embeddings[1][i][1].split("/")[:-3]))
return model, path_images, embeddings
print("Cargando modelos...")
model, path_images, embeddings = load()
source = {0: "\nDocExplore SAM", 1: "\nDocExplore GroundingDINO"}
stroke_width = st.sidebar.slider("Stroke width: ", 1, 25, 5)
dataset_transforms = transforms.Compose([
transforms.Resize((224, 224)),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
])
def compute_text_embeddings(sketch):
with torch.no_grad():
sketch_feat = model(sketch.to(device), dtype='sketch')
return sketch_feat
# inputs = clip_processor(text=list_of_strings, return_tensors="pt", padding=True).to(
# device
# )
# with torch.no_grad():
# result = clip_model.get_text_features(**inputs).detach().cpu().numpy()
# return result / np.linalg.norm(result, axis=1, keepdims=True)
#return torch.randn(1, 768)
def image_search(query, corpus, n_results=N_RESULTS):
query_embedding = compute_text_embeddings(query)
corpus_id = 0 if corpus == "DocExplore SAM" else 1
image_features = torch.tensor([item[0] for item in embeddings[corpus_id]]).to(device)
bbox_of_images = torch.tensor([item[1] for item in embeddings[corpus_id]]).to(device)
label_of_images = torch.tensor([item[2] for item in embeddings[corpus_id]]).to(device)
dot_product = (image_features @ query_embedding.T)[:, 0]
_, max_indices = torch.topk(dot_product, n_results, dim=0, largest=True, sorted=True)
return [
(
path_images + "page" + str(i) + ".jpg",
)
for i in label_of_images[max_indices].cpu().numpy().tolist()
], bbox_of_images[max_indices], dot_product[max_indices]
def make_square(img, fill_color=(255, 255, 255)):
x, y = img.size
size = max(x, y)
new_img = Image.new("RGB", (x, y), fill_color)
new_img.paste(img)
return new_img, x, y
@st.cache_data
def get_images(paths):
def process_image(path):
return make_square(Image.open(path))
processed = Pool(N_RESULTS).map(process_image, paths)
imgs, xs, ys = [], [], []
for img, x, y in processed:
imgs.append(img)
xs.append(x)
ys.append(y)
return imgs, xs, ys
def keep_best_boxes(boxes, scores, score_threshold=0.1, max_iou=0.8):
candidates = []
for box, score in zip(boxes, scores):
box = [round(i, 0) for i in box.tolist()]
if score >= score_threshold:
candidates.append((box, float(score)))
to_ignore = set()
for i in range(len(candidates) - 1):
if i in to_ignore:
continue
for j in range(i + 1, len(candidates)):
if j in to_ignore:
continue
xmin1, ymin1, xmax1, ymax1 = candidates[i][0]
xmin2, ymin2, xmax2, ymax2 = candidates[j][0]
if xmax1 < xmin2 or xmax2 < xmin1 or ymax1 < ymin2 or ymax2 < ymin1:
continue
else:
xmin_inter, xmax_inter = sorted(
[xmin1, xmax1, xmin2, xmax2])[1:3]
ymin_inter, ymax_inter = sorted(
[ymin1, ymax1, ymin2, ymax2])[1:3]
area_inter = (xmax_inter - xmin_inter) * \
(ymax_inter - ymin_inter)
area1 = (xmax1 - xmin1) * (ymax1 - ymin1)
area2 = (xmax2 - xmin2) * (ymax2 - ymin2)
iou = area_inter / (area1 + area2 - area_inter)
if iou > max_iou:
if candidates[i][1] > candidates[j][1]:
to_ignore.add(j)
else:
to_ignore.add(i)
break
else:
if area_inter / area1 > 0.9:
if candidates[i][1] < 1.1 * candidates[j][1]:
to_ignore.add(i)
if area_inter / area2 > 0.9:
if 1.1 * candidates[i][1] > candidates[j][1]:
to_ignore.add(j)
return [candidates[i][0] for i in range(len(candidates)) if i not in to_ignore]
def convert_pil_to_base64(image):
img_buffer = BytesIO()
image.save(img_buffer, format="JPEG")
byte_data = img_buffer.getvalue()
base64_str = base64.b64encode(byte_data)
return base64_str
def draw_reshape_encode(img, boxes, x, y):
boxes = [boxes.tolist()]
image = img.copy()
draw = ImageDraw.Draw(image)
new_x, new_y = int(x * HEIGHT / y), HEIGHT
for box in boxes:
print("box:", box)
draw.rectangle(
[(box[0], box[1]), (box[2], box[3])], # (x_min, y_min, x_max, y_max)
outline=color, # Box color
width=10 # Box width
)
#if x > y:
# image = image.crop((0, (x - y) / 2, x, x - (x - y) / 2))
#else:
# image = image.crop(((y - x) / 2, 0, y - (y - x) / 2, y))
return convert_pil_to_base64(image.resize((new_x, new_y)))
def get_html(url_list, encoded_images):
html = "<div style='margin-top: 20px; max-width: 1200px; display: flex; flex-wrap: wrap; justify-content: space-evenly'>"
for i in range(len(url_list)):
title, encoded = url_list[i][0], encoded_images[i]
html = (
html
+ f"<img title='{escape(title)}' style='height: {HEIGHT}px; margin: 1px' src='data:image/jpeg;base64,{encoded.decode()}'>"
)
html += "</div>"
return html
description = """
# Sketch-based Detection
This app retrieves images from the [DocExplore](https://www.docexplore.eu/?lang=en) dataset based on a sketch query.
**Tip 1**: you can draw a sketch in the canvas.
**Tip 2**: you can change the size of the stroke with the slider.
The model utilized in this application is a DINOv2, which was trained in a self-supervised manner on the Flickr25k dataset.
"""
div_style = {
"display": "flex",
"justify-content": "center",
"flex-wrap": "wrap",
}
def main():
st.markdown(
"""
<style>
.block-container{
max-width: 1600px;
}
div.row-widget > div{
flex-direction: row;
display: flex;
justify-content: center;
}
div.row-widget.stRadio > div > label{
margin-left: 5px;
margin-right: 5px;
}
.row-widget {
margin-top: -25px;
}
section > div:first-child {
padding-top: 30px;
}
div.appview-container > section:first-child{
max-width: 320px;
}
#MainMenu {
visibility: hidden;
}
.stMarkdown {
display: grid;
place-items: center;
}
</style>
""",
unsafe_allow_html=True,
)
st.sidebar.markdown(description)
st.title("One-Shot Detection")
# Create two main columns
left_col, right_col = st.columns([0.2, 0.8]) # Adjust the weights as needed
with left_col:
# Canvas for drawing
canvas_result = st_canvas(
background_color="#eee",
stroke_width=stroke_width,
update_streamlit=True,
height=300,
width=300,
key="color_annotation_app",
)
# Input controls
query = [0]
corpus = st.radio("", ["DocExplore SAM", "DocExplore GroundingDINO"], index=0)
# score_threshold = st.slider(
# "Score threshold", min_value=0.01, max_value=1.0, value=0.5, step=0.01
# )
with right_col:
if canvas_result.image_data is not None:
draw = Image.fromarray(canvas_result.image_data.astype("uint8"))
draw = ImageOps.pad(draw.convert("RGB"), size=(224, 224))
draw.save("draw.jpg")
draw_tensor = transforms.ToTensor()(draw)
draw_tensor = transforms.Resize((224, 224))(draw_tensor)
draw_tensor = transforms.Normalize(
mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]
)(draw_tensor)
draw_tensor = draw_tensor.unsqueeze(0)
else:
return
if len(query) > 0:
retrieved, bbox_of_images, dot_product = image_search(draw_tensor, corpus)
imgs, xs, ys = get_images([x[0] for x in retrieved])
encoded_images = []
for image_idx in range(len(imgs)):
img0, x, y = imgs[image_idx], xs[image_idx], ys[image_idx]
encoded_images.append(draw_reshape_encode(img0, bbox_of_images[image_idx], x, y))
st.markdown(get_html(retrieved, encoded_images), unsafe_allow_html=True)
if __name__ == "__main__":
main()