gradio / cxr_gradio.py
eprakash's picture
Upload folder using huggingface_hub
711ffc5 verified
import gradio as gr
import random
import time
import os
from glob import glob
from PIL import Image
import torchvision.transforms as transforms
image_prefix = "/deep/u/eprakash/AngioSeg/diffusion/cxr_synthetic_data_25_no_transform/synth/"
image_ids = glob(os.path.join(image_prefix, '*' + '.png'))
image_ids = list(set([os.path.splitext(os.path.basename(p))[0].split("_")[0] for p in image_ids]))
save_path = "cxr_ranks"
def load_img(img_path, size=1024):
img = Image.open(img_path).convert('RGB')
transform_list = [transforms.Resize((size, size))]
transform = transforms.Compose(transform_list)
img = transform(img)
return img
def find_completed_idxs(save_path=save_path):
files = os.listdir(save_path)
if len(files) == 0:
return [-1]
else:
file_list = []
for f in files:
f = int(f.split(".")[0])
file_list.append(f)
file_list = sorted(file_list)
return file_list
def load_next(rank, img_1, mask_1, img_2, mask_2, img_3, mask_3, img_4, mask_4, example, ids=image_ids, image_prefix=image_prefix, save_path=save_path):
if int(example) == len(image_ids) - 1:
return [None, None, None, None, None, None, None]
else:
file_list = find_completed_idxs()
if (int(example) not in file_list):
r = str(image_ids[int(example)]) + "," + rank
r_fp = open(save_path + "/" + str(int(example)) +".txt", "w")
r_fp.write(r + "\n")
r_fp.close()
file_list = find_completed_idxs()
example = file_list[-1] + 1
rank = ""
img_1 = gr.Image(label="Sample #1", value=load_img(image_prefix + str(image_ids[int(example)]) + "_synthetic_0.png"), interactive=False)
mask_1 = gr.Image(label="Mask", value=load_img(image_prefix + str(image_ids[int(example)]) + "_synthetic_mask_1.png"), interactive=False)
img_2 = gr.Image(label="Sample #2", value=load_img(image_prefix+ str(image_ids[int(example)]) + "_synthetic_1.png"), interactive=False)
mask_2 = gr.Image(label="Mask", value=load_img(image_prefix + str(image_ids[int(example)]) + "_synthetic_mask_2.png"), interactive=False)
img_3 = gr.Image(label="Sample #3", value=load_img(image_prefix + str(image_ids[int(example)]) + "_synthetic_2.png"), interactive=False)
mask_3 = gr.Image(label="Mask", value=load_img(image_prefix + str(image_ids[int(example)]) + "_synthetic_mask_3.png"), interactive=False)
img_4 = gr.Image(label="Sample #4", value=load_img(image_prefix + str(image_ids[int(example)]) + "_synthetic_3.png"), interactive=False)
mask_4 = gr.Image(label="Mask", value=load_img(image_prefix + str(image_ids[int(example)]) + "_synthetic_mask_4.png"), interactive=False)
return [rank, img_1, mask_1, img_2, mask_2, img_3, mask_3, img_4, mask_4, example]
with gr.Blocks() as demo:
last_idx = -1
example = gr.Number(label="Example #. Click next for #-1 (blank starting page).", value=last_idx, interactive=False)
rank = gr.Textbox(label="Rankings (Best to worst, comma-separated, no spaces).")
with gr.Column(scale=1):
with gr.Row():
mask_1 = gr.Image(label="Mask", value=load_img("/deep/u/eprakash/blank.jpg"), interactive=False)
img_1 = gr.Image(label="Sample #1", value=load_img("/deep/u/eprakash/blank.jpg"), interactive=False)
with gr.Row():
mask_2 = gr.Image(label="Mask", value=load_img("/deep/u/eprakash/blank.jpg"), interactive=False)
img_2 = gr.Image(label="Sample #2", value=load_img("/deep/u/eprakash/blank.jpg"), interactive=False)
with gr.Row():
mask_3 = gr.Image(label="Mask", value=load_img("/deep/u/eprakash/blank.jpg"), interactive=False)
img_3 = gr.Image(label="Sample #3", value=load_img("/deep/u/eprakash/blank.jpg"), interactive=False)
with gr.Row():
mask_4 = gr.Image(label="Mask", value=load_img("/deep/u/eprakash/blank.jpg"), interactive=False)
img_4 = gr.Image(label="Sample #4", value=load_img("/deep/u/eprakash/blank.jpg"), interactive=False)
next_btn = gr.Button(value="Next")
next_btn.click(fn=load_next, inputs=[rank, img_1, mask_1, img_2, mask_2, img_3, mask_3, img_4, mask_4, example], outputs=[rank, img_1, mask_1, img_2, mask_2, img_3, mask_3, img_4, mask_4, example], queue=False)
demo.queue()
demo.launch(share=True)