Spaces:

noelshin
/

selfmask

Runtime error

App Files Files Community

noelshin commited on Jun 16, 2022

Commit

7b03ec2

•

1 Parent(s): 68aa8af

downgrade gradio to 2.9.0 from 3.0.17 for a better example UI

Browse files

Files changed (4) hide show

README.md +2 -2
__pycache__/bilateral_solver.cpython-38.pyc +0 -0
app.py +12 -26
description.html +9 -3

README.md CHANGED Viewed

@@ -1,10 +1,10 @@
 ---
 title: Selfmask
-emoji: 📉
 colorFrom: gray
 colorTo: gray
 sdk: gradio
-sdk_version: 3.0.17
 app_file: app.py
 pinned: false
 license: mit

 ---
 title: Selfmask
+emoji: 😷
 colorFrom: gray
 colorTo: gray
 sdk: gradio
+sdk_version: 2.9.0
 app_file: app.py
 pinned: false
 license: mit

__pycache__/bilateral_solver.cpython-38.pyc CHANGED Viewed

Binary files a/__pycache__/bilateral_solver.cpython-38.pyc and b/__pycache__/bilateral_solver.cpython-38.pyc differ

app.py CHANGED Viewed

@@ -1,5 +1,6 @@
 from argparse import ArgumentParser, Namespace
 from typing import Dict, List, Tuple
 import yaml
 import numpy as np
 import cv2
@@ -26,23 +27,6 @@ parser.add_argument(
     default="duts-dino-k234-nq20-224-swav-mocov2-dino-p16-sr10100.yaml"
 )
-# parser.add_argument(
-#     "--p_state_dict",
-#     type=str,
-#     default="/users/gyungin/selfmask_bak/ckpt/nq20_ndl6_bc_sr10100_duts_pm_all_k2,3,4_md_seed0_final/eval/hku_is/best_model.pt",
-# )
-#
-# parser.add_argument(
-#     "--dataset_name", '-dn', type=str, default="duts",
-#     choices=["dut_omron", "duts", "ecssd"]
-# )
-# independent variables
-# parser.add_argument("--use_gpu", type=bool, default=True)
-# parser.add_argument('--seed', default=0, type=int)
-# parser.add_argument("--dir_root", type=str, default="..")
-# parser.add_argument("--gpu_id", type=int, default=2)
-# parser.add_argument("--suffix", type=str, default='')
 args: Namespace = parser.parse_args()
 base_args = yaml.safe_load(open(f"{args.config}", 'r'))
 base_args.pop("dataset_name")
@@ -54,15 +38,14 @@ model = get_model(arch="maskformer", configs=args).to(device)
 model.load_state_dict(state_dict)
 model.eval()
 @torch.no_grad()
-def main(
-        image: Image.Image,
-        size: int = 384,
-        max_size: int = 512,
-        mean: Tuple[float, float, float] = (0.485, 0.456, 0.406),
-        std: Tuple[float, float, float] = (0.229, 0.224, 0.225)
-):
     pil_image: Image.Image = resize(image, size=size, max_size=max_size)
     image: torch.Tensor = normalize(to_tensor(pil_image), mean=list(mean), std=list(std))  # 3 x H x W
     dict_outputs = model(image[None].to(device))
@@ -101,10 +84,11 @@ def main(
     return super_imposed_img
     # return pred_mask_bi
 demo = gr.Interface(
     fn=main,
-    inputs=gr.inputs.Image(type="pil"),
-    outputs="image",
     examples=[f"resources/{fname}.jpg" for fname in [
         "0053",
         "0236",
@@ -124,6 +108,8 @@ demo = gr.Interface(
         "sun_amnrcxhisjfrliwa",
         "sun_bvyxpvkouzlfwwod"
     ]],
     title="Unsupervised Salient Object Detection with Spectral Cluster Voting",
     allow_flagging="never",
     analytics_enabled=False

 from argparse import ArgumentParser, Namespace
 from typing import Dict, List, Tuple
+import codecs
 import yaml
 import numpy as np
 import cv2
     default="duts-dino-k234-nq20-224-swav-mocov2-dino-p16-sr10100.yaml"
 )
 args: Namespace = parser.parse_args()
 base_args = yaml.safe_load(open(f"{args.config}", 'r'))
 base_args.pop("dataset_name")
 model.load_state_dict(state_dict)
 model.eval()
+size: int = 384
+max_size: int = 512
+mean: Tuple[float, float, float] = (0.485, 0.456, 0.406)
+std: Tuple[float, float, float] = (0.229, 0.224, 0.225)
 @torch.no_grad()
+def main(image: Image):
     pil_image: Image.Image = resize(image, size=size, max_size=max_size)
     image: torch.Tensor = normalize(to_tensor(pil_image), mean=list(mean), std=list(std))  # 3 x H x W
     dict_outputs = model(image[None].to(device))
     return super_imposed_img
     # return pred_mask_bi
 demo = gr.Interface(
     fn=main,
+    inputs=gr.inputs.Image(type="pil", source="upload", tool="editor"),
+    outputs=gr.outputs.Image(type="numpy", label="saliency map"),  # "image",
     examples=[f"resources/{fname}.jpg" for fname in [
         "0053",
         "0236",
         "sun_amnrcxhisjfrliwa",
         "sun_bvyxpvkouzlfwwod"
     ]],
+    examples_per_page=20,
+    description=codecs.open("description.html", 'r', "utf-8").read(),
     title="Unsupervised Salient Object Detection with Spectral Cluster Voting",
     allow_flagging="never",
     analytics_enabled=False

description.html CHANGED Viewed

@@ -1,10 +1,16 @@
 <!DOCTYPE html>
 <html lang="en">
 <head>
-  <meta charset="UTF-8">
-  <title>$Title$</title>
 </head>
 <body>
-$END$
 </body>
 </html>

 <!DOCTYPE html>
 <html lang="en">
 <head>
+    <meta charset="UTF-8">
+    <title>Title</title>
 </head>
 <body>
+    This is a demo of <a href="https://arxiv.org/pdf/2203.12614.pdf">Unsupervised Salient Object Detection with Spectral Cluster Voting</a> (CVPRW 2022).</br>
+    In the paper, we tackle the challenging task of unsupervised salient object detection (SOD) by leveraging spectral clustering on self-supervised features.
+    We make the following contributions:
+    (i) We revisit spectral clustering and demonstrate its potential to group the pixels of salient objects;
+    (ii) Given mask proposals from multiple applications of spectral clustering on image features computed from various self-supervised models, e.g., MoCov2, SwAV, DINO, we propose a simple but effective winner-takes-all voting mechanism for selecting the salient masks, leveraging object priors based on framing and distinctiveness;
+    (iii) Using the selected object segmentation as pseudo groundtruth masks, we train a salient object detector, dubbed SelfMask, which outperforms prior approaches on three unsupervised SOD benchmarks.
+    Code is publicly available at <a href="https://github.com/NoelShin/selfmask">our repo</a>.
 </body>
 </html>