Spaces:
Runtime error
Runtime error
edadaltocg
commited on
Commit
•
df37aa3
1
Parent(s):
db83f58
update images
Browse files- app.py +48 -27
- images/imagenet/n02828884_603_bench.jpg +0 -0
- images/imagenet/n02834778_3678_bicycle.jpg +0 -0
- images/imagenet/n02880940_17711_bowl.jpg +0 -0
- images/imagenet/n03062245_2005_cocktail_shaker.jpg +0 -0
- images/imagenet/n03495258_9079_harp.jpg +0 -0
- images/ood/Rademacher_025_Rademacher_02897.jpg +0 -0
- images/ood/art_2.jpg +0 -0
- images/ood/bumpy_0140.jpg +0 -0
- images/ood/door_022_00033.jpg +0 -0
- images/ood/fdb9d2ac3f37c0c80baa7f91775e58ce.jpg +0 -0
- images/ood/fed8bd31654ee16a9cd83c8de72ddb5b.jpg +0 -0
- images/ood/ff7f83dfb2485306b62bf64726f4f932.jpg +0 -0
- images/ood/ffd5b90b142ebcb46cffc96314e6bcd3.jpg +0 -0
- images/ood/fireworks_001_0001.png +0 -0
- images/ood/i_ice_floe_00002019.jpg +0 -0
- images/ood/i_igloo_00002495.jpg +0 -0
- images/ood/knitted_0141.jpg +0 -0
- images/ood/pyramid_008_image_0011.jpg +0 -0
- images/ood/scissors_040_scissors_0085_pixabay.jpg +0 -0
- images/ood/striped_0063.jpg +0 -0
- images/ood/sun_awovauomdhnolaul.jpg +0 -0
- images/ood/sun_bzrmbfcxyebbxuqu.jpg +0 -0
- images/ood/sun_bzuroamlnffhyuqn.jpg +0 -0
- images/ood/toy_2.jpg +0 -0
- images/ood/w_waterfall_00004924.jpg +0 -0
- images/ood/w_wheat_field_00004628.jpg +0 -0
- images/ood/watermelon_0.9992305.JPEG +0 -0
app.py
CHANGED
@@ -3,22 +3,21 @@ Gradio demo of image classification with OOD detection.
|
|
3 |
|
4 |
If the image example is probably OOD, the model will abstain from the prediction.
|
5 |
"""
|
6 |
-
import os
|
7 |
-
import pickle
|
8 |
import json
|
|
|
|
|
9 |
from glob import glob
|
10 |
|
11 |
import gradio as gr
|
12 |
-
from gradio.components import Image, Label, JSON
|
13 |
import numpy as np
|
14 |
-
import torch
|
15 |
import timm
|
|
|
|
|
|
|
16 |
from timm.data import resolve_data_config
|
17 |
from timm.data.transforms_factory import create_transform
|
18 |
from torchvision.models.feature_extraction import create_feature_extractor, get_graph_node_names
|
19 |
|
20 |
-
import logging
|
21 |
-
|
22 |
_logger = logging.getLogger(__name__)
|
23 |
|
24 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
@@ -26,7 +25,7 @@ TOPK = 3
|
|
26 |
|
27 |
# load model
|
28 |
print("Loading model...")
|
29 |
-
model = timm.create_model("resnet50", pretrained=True)
|
30 |
model.to(device)
|
31 |
model.eval()
|
32 |
|
@@ -34,28 +33,25 @@ model.eval()
|
|
34 |
idx2label = json.loads(open("ilsvrc2012.json").read())
|
35 |
idx2label = {int(k): v for k, v in idx2label.items()}
|
36 |
print(idx2label)
|
|
|
37 |
|
38 |
# transformation
|
39 |
config = resolve_data_config({}, model=model)
|
40 |
config["is_training"] = False
|
41 |
transform = create_transform(**config)
|
42 |
|
43 |
-
#
|
44 |
-
print(get_graph_node_names(model)[0])
|
45 |
-
|
46 |
-
# load train scores
|
47 |
penultimate_features_key = "global_pool.flatten"
|
48 |
logits_key = "fc"
|
49 |
features_names = [penultimate_features_key, logits_key]
|
50 |
|
51 |
-
# create feature extractor
|
52 |
feature_extractor = create_feature_extractor(model, features_names)
|
53 |
|
54 |
-
|
|
|
55 |
msp_threshold = 0.3796
|
56 |
energy_threshold = 0.3781
|
57 |
-
|
58 |
-
## unpickle detectors
|
59 |
|
60 |
|
61 |
def mahalanobis_penult(features):
|
@@ -72,9 +68,18 @@ def energy(logits):
|
|
72 |
return torch.logsumexp(logits, dim=1).item()
|
73 |
|
74 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
75 |
def predict(image):
|
76 |
# forward pass
|
77 |
inputs = transform(image).unsqueeze(0)
|
|
|
78 |
with torch.no_grad():
|
79 |
features = feature_extractor(inputs)
|
80 |
|
@@ -86,13 +91,16 @@ def predict(image):
|
|
86 |
|
87 |
result = {idx2label[i.item()]: v.item() for i, v in zip(class_idxs.squeeze(), softmax.squeeze())}
|
88 |
# OOD
|
89 |
-
msp_score = msp(features[logits_key])
|
90 |
-
energy_score = energy(features[logits_key])
|
|
|
91 |
ood_scores = {
|
92 |
-
"
|
93 |
-
"
|
94 |
-
"
|
95 |
-
"
|
|
|
|
|
96 |
}
|
97 |
_logger.info(ood_scores)
|
98 |
return result, ood_scores
|
@@ -100,9 +108,9 @@ def predict(image):
|
|
100 |
|
101 |
def main():
|
102 |
# image examples for demo shuffled
|
103 |
-
examples = glob("images/imagenet
|
104 |
np.random.seed(42)
|
105 |
-
np.random.shuffle(examples)
|
106 |
|
107 |
# gradio interface
|
108 |
interface = gr.Interface(
|
@@ -117,11 +125,24 @@ def main():
|
|
117 |
allow_flagging="never",
|
118 |
theme="default",
|
119 |
title="OOD Detection 🧐",
|
120 |
-
description=
|
121 |
-
|
122 |
-
|
123 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
124 |
)
|
|
|
125 |
interface.close()
|
126 |
|
127 |
|
|
|
3 |
|
4 |
If the image example is probably OOD, the model will abstain from the prediction.
|
5 |
"""
|
|
|
|
|
6 |
import json
|
7 |
+
import logging
|
8 |
+
import pickle
|
9 |
from glob import glob
|
10 |
|
11 |
import gradio as gr
|
|
|
12 |
import numpy as np
|
|
|
13 |
import timm
|
14 |
+
import torch
|
15 |
+
import torch.nn.functional as F
|
16 |
+
from gradio.components import JSON, Image, Label
|
17 |
from timm.data import resolve_data_config
|
18 |
from timm.data.transforms_factory import create_transform
|
19 |
from torchvision.models.feature_extraction import create_feature_extractor, get_graph_node_names
|
20 |
|
|
|
|
|
21 |
_logger = logging.getLogger(__name__)
|
22 |
|
23 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
|
|
25 |
|
26 |
# load model
|
27 |
print("Loading model...")
|
28 |
+
model = timm.create_model("resnet50.tv2_in1k", pretrained=True, checkpoint_path="resnet50.tv2_in1k.bin")
|
29 |
model.to(device)
|
30 |
model.eval()
|
31 |
|
|
|
33 |
idx2label = json.loads(open("ilsvrc2012.json").read())
|
34 |
idx2label = {int(k): v for k, v in idx2label.items()}
|
35 |
print(idx2label)
|
36 |
+
print(idx2label.values())
|
37 |
|
38 |
# transformation
|
39 |
config = resolve_data_config({}, model=model)
|
40 |
config["is_training"] = False
|
41 |
transform = create_transform(**config)
|
42 |
|
43 |
+
# create feature extractor
|
|
|
|
|
|
|
44 |
penultimate_features_key = "global_pool.flatten"
|
45 |
logits_key = "fc"
|
46 |
features_names = [penultimate_features_key, logits_key]
|
47 |
|
|
|
48 |
feature_extractor = create_feature_extractor(model, features_names)
|
49 |
|
50 |
+
centroids = torch.from_numpy(pickle.load(open("centroids_resnet50.tv2_in1k_igeood_logits.pkl", "rb"))).to(device)
|
51 |
+
# OOD detector thresholds
|
52 |
msp_threshold = 0.3796
|
53 |
energy_threshold = 0.3781
|
54 |
+
igeood_threshold = 2.4984
|
|
|
55 |
|
56 |
|
57 |
def mahalanobis_penult(features):
|
|
|
68 |
return torch.logsumexp(logits, dim=1).item()
|
69 |
|
70 |
|
71 |
+
def igeoodlogits_vec(logits, temperature, centroids, epsilon=1e-12):
|
72 |
+
logits = torch.sqrt(F.softmax(logits / temperature, dim=1))
|
73 |
+
centroids = torch.sqrt(F.softmax(centroids / temperature, dim=1))
|
74 |
+
mult = logits @ centroids.T
|
75 |
+
stack = 2 * torch.acos(torch.clamp(mult, -1 + epsilon, 1 - epsilon))
|
76 |
+
return stack.mean(dim=1).item()
|
77 |
+
|
78 |
+
|
79 |
def predict(image):
|
80 |
# forward pass
|
81 |
inputs = transform(image).unsqueeze(0)
|
82 |
+
inputs = inputs.to(device)
|
83 |
with torch.no_grad():
|
84 |
features = feature_extractor(inputs)
|
85 |
|
|
|
91 |
|
92 |
result = {idx2label[i.item()]: v.item() for i, v in zip(class_idxs.squeeze(), softmax.squeeze())}
|
93 |
# OOD
|
94 |
+
msp_score = round(msp(features[logits_key]), 4)
|
95 |
+
energy_score = round(energy(features[logits_key]), 4)
|
96 |
+
igeood_scores = round(igeoodlogits_vec(features[logits_key], 1, centroids), 4)
|
97 |
ood_scores = {
|
98 |
+
"MSP": msp_score,
|
99 |
+
"MSP, is the input OOD?": msp_score < msp_threshold,
|
100 |
+
"Energy": energy_score,
|
101 |
+
"Energy, is the input OOD?": energy_score < energy_threshold,
|
102 |
+
"Igeood": igeood_scores,
|
103 |
+
"Igeood, is the input OOD?": igeood_scores < igeood_threshold,
|
104 |
}
|
105 |
_logger.info(ood_scores)
|
106 |
return result, ood_scores
|
|
|
108 |
|
109 |
def main():
|
110 |
# image examples for demo shuffled
|
111 |
+
examples = glob("images/imagenet/*") + glob("images/ood/*")
|
112 |
np.random.seed(42)
|
113 |
+
# np.random.shuffle(examples)
|
114 |
|
115 |
# gradio interface
|
116 |
interface = gr.Interface(
|
|
|
125 |
allow_flagging="never",
|
126 |
theme="default",
|
127 |
title="OOD Detection 🧐",
|
128 |
+
description=(
|
129 |
+
"Out-of-distribution (OOD) detection is an essential safety measure for machine learning models. "
|
130 |
+
"The objective of an OOD detector is to determine wether the input sample comes from the distribution known by the AI model. "
|
131 |
+
"For instance, an input that does not belong to any of the known classes or is from a different domain should be flagged by the detector.\n"
|
132 |
+
"In this demo we will display the decision of three OOD detectors on a ResNet-50 model trained to classify on the ImageNet-1K dataset (top-1 accuracy 80%)."
|
133 |
+
"This model can classify among 1000 classes from several categories, including `animals`, `vehicles`, `clothing`, `instruments`, `plants`, etc. "
|
134 |
+
"For the complete hierarchy of classes, please check the website https://observablehq.com/@mbostock/imagenet-hierarchy. "
|
135 |
+
"\n\n"
|
136 |
+
"## Instructions:\n"
|
137 |
+
"1. Upload an image of your choice or select one from the examples bar.\n"
|
138 |
+
"2. The model will predict the top 3 most likely classes for the image.\n"
|
139 |
+
"3. The OOD detectors will output their scores and decision on the image. The smaller the score, the least confident the detector is on the sample being in-distribution.\n"
|
140 |
+
"4. If the image is OOD, the model will abstain from the prediction and flag it to the practicioner.\n"
|
141 |
+
"\n\n\nEnjoy the demo!"
|
142 |
+
),
|
143 |
+
cache_examples=True,
|
144 |
)
|
145 |
+
interface.launch(server_port=7860)
|
146 |
interface.close()
|
147 |
|
148 |
|
images/imagenet/n02828884_603_bench.jpg
DELETED
Binary file (148 kB)
|
|
images/imagenet/n02834778_3678_bicycle.jpg
DELETED
Binary file (75.3 kB)
|
|
images/imagenet/n02880940_17711_bowl.jpg
DELETED
Binary file (68.9 kB)
|
|
images/imagenet/n03062245_2005_cocktail_shaker.jpg
DELETED
Binary file (57.7 kB)
|
|
images/imagenet/n03495258_9079_harp.jpg
DELETED
Binary file (112 kB)
|
|
images/ood/Rademacher_025_Rademacher_02897.jpg
ADDED
images/ood/art_2.jpg
ADDED
images/ood/bumpy_0140.jpg
DELETED
Binary file (105 kB)
|
|
images/ood/door_022_00033.jpg
ADDED
images/ood/fdb9d2ac3f37c0c80baa7f91775e58ce.jpg
DELETED
Binary file (577 kB)
|
|
images/ood/fed8bd31654ee16a9cd83c8de72ddb5b.jpg
DELETED
Binary file (752 kB)
|
|
images/ood/ff7f83dfb2485306b62bf64726f4f932.jpg
DELETED
Binary file (191 kB)
|
|
images/ood/ffd5b90b142ebcb46cffc96314e6bcd3.jpg
DELETED
Binary file (206 kB)
|
|
images/ood/fireworks_001_0001.png
ADDED
images/ood/i_ice_floe_00002019.jpg
DELETED
Binary file (58.3 kB)
|
|
images/ood/i_igloo_00002495.jpg
DELETED
Binary file (60.6 kB)
|
|
images/ood/knitted_0141.jpg
DELETED
Binary file (71.9 kB)
|
|
images/ood/pyramid_008_image_0011.jpg
ADDED
images/ood/scissors_040_scissors_0085_pixabay.jpg
ADDED
images/ood/striped_0063.jpg
DELETED
Binary file (22.3 kB)
|
|
images/ood/sun_awovauomdhnolaul.jpg
DELETED
Binary file (33.2 kB)
|
|
images/ood/sun_bzrmbfcxyebbxuqu.jpg
DELETED
Binary file (361 kB)
|
|
images/ood/sun_bzuroamlnffhyuqn.jpg
DELETED
Binary file (170 kB)
|
|
images/ood/toy_2.jpg
ADDED
images/ood/w_waterfall_00004924.jpg
DELETED
Binary file (69.3 kB)
|
|
images/ood/w_wheat_field_00004628.jpg
DELETED
Binary file (105 kB)
|
|
images/ood/watermelon_0.9992305.JPEG
ADDED