Spaces:
Runtime error
Runtime error
format changes
Browse files
app.py
CHANGED
@@ -11,17 +11,19 @@ from msma import ScoreFlow, config_presets
|
|
11 |
|
12 |
|
13 |
@cache
|
14 |
-
def load_model(modeldir, preset="edm2-img64-s-fid", device=
|
15 |
model = ScoreFlow(preset, device=device)
|
16 |
model.flow.load_state_dict(torch.load(f"{modeldir}/{preset}/flow.pt"))
|
17 |
return model
|
18 |
|
|
|
19 |
@cache
|
20 |
def load_reference_scores(model_dir):
|
21 |
with np.load(f"{model_dir}/refscores.npz", "rb") as f:
|
22 |
ref_nll = f["arr_0"]
|
23 |
return ref_nll
|
24 |
|
|
|
25 |
def compute_gmm_likelihood(x_score, model_dir):
|
26 |
with open(f"{model_dir}/gmm.pkl", "rb") as f:
|
27 |
clf = load(f)
|
@@ -32,47 +34,53 @@ def compute_gmm_likelihood(x_score, model_dir):
|
|
32 |
|
33 |
return nll, percentile, ref_nll
|
34 |
|
|
|
35 |
def plot_against_reference(nll, ref_nll):
|
36 |
fig, ax = plt.subplots()
|
37 |
ax.hist(ref_nll, label="Reference Scores")
|
38 |
-
ax.axvline(nll, label=
|
39 |
plt.legend()
|
40 |
fig.tight_layout()
|
41 |
return fig
|
42 |
|
|
|
43 |
def plot_heatmap(img: Image, heatmap: np.array):
|
44 |
fig, ax = plt.subplots()
|
45 |
cmap = plt.get_cmap("gist_heat")
|
46 |
-
h = heatmap[0,0].copy()
|
47 |
-
qmin, qmax = np.quantile(h, 0.
|
48 |
h = np.clip(h, a_min=qmin, a_max=qmax)
|
49 |
-
h = (h-h.min()) / (h.max() - h.min())
|
50 |
-
h = cmap(h, bytes=True)[
|
51 |
h = Image.fromarray(h).resize(img.size, resample=Image.Resampling.BILINEAR)
|
52 |
im = Image.blend(img, h, alpha=0.6)
|
53 |
-
im = ax.imshow(np.array(im))
|
54 |
-
# fig.colorbar(im)
|
55 |
-
# plt.grid(False)
|
56 |
-
# plt.axis("off")
|
57 |
-
fig.tight_layout()
|
58 |
-
return
|
59 |
|
60 |
-
|
|
|
61 |
|
62 |
# img = center_crop_imagenet(64, img)
|
63 |
input_img = input_img.resize(size=(64, 64), resample=Image.Resampling.LANCZOS)
|
64 |
|
65 |
with torch.inference_mode():
|
66 |
img = np.array(input_img)
|
67 |
-
img = torch.from_numpy(img).permute(2,0,1).unsqueeze(0)
|
68 |
img = img.float().to(device)
|
69 |
-
model = load_model(modeldir=
|
70 |
img_likelihood = model(img).cpu().numpy()
|
71 |
-
|
72 |
-
|
|
|
73 |
x = model.scorenet(img)
|
74 |
x = x.square().sum(dim=(2, 3, 4)) ** 0.5
|
75 |
-
nll, pct, ref_nll = compute_gmm_likelihood(
|
|
|
|
|
76 |
|
77 |
outstr = f"Anomaly score: {nll:.3f} / {pct:.2f} percentile"
|
78 |
histplot = plot_against_reference(nll, ref_nll)
|
@@ -83,11 +91,19 @@ def run_inference(input_img, preset="edm2-img64-s-fid", device="cuda"):
|
|
83 |
|
84 |
demo = gr.Interface(
|
85 |
fn=run_inference,
|
86 |
-
inputs=[
|
87 |
-
|
88 |
-
|
89 |
-
|
90 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
91 |
)
|
92 |
|
93 |
if __name__ == "__main__":
|
|
|
11 |
|
12 |
|
13 |
@cache
|
14 |
+
def load_model(modeldir, preset="edm2-img64-s-fid", device="cpu", outdir=None):
|
15 |
model = ScoreFlow(preset, device=device)
|
16 |
model.flow.load_state_dict(torch.load(f"{modeldir}/{preset}/flow.pt"))
|
17 |
return model
|
18 |
|
19 |
+
|
20 |
@cache
|
21 |
def load_reference_scores(model_dir):
|
22 |
with np.load(f"{model_dir}/refscores.npz", "rb") as f:
|
23 |
ref_nll = f["arr_0"]
|
24 |
return ref_nll
|
25 |
|
26 |
+
|
27 |
def compute_gmm_likelihood(x_score, model_dir):
|
28 |
with open(f"{model_dir}/gmm.pkl", "rb") as f:
|
29 |
clf = load(f)
|
|
|
34 |
|
35 |
return nll, percentile, ref_nll
|
36 |
|
37 |
+
|
38 |
def plot_against_reference(nll, ref_nll):
|
39 |
fig, ax = plt.subplots()
|
40 |
ax.hist(ref_nll, label="Reference Scores")
|
41 |
+
ax.axvline(nll, label="Image Score", c="red", ls="--")
|
42 |
plt.legend()
|
43 |
fig.tight_layout()
|
44 |
return fig
|
45 |
|
46 |
+
|
47 |
def plot_heatmap(img: Image, heatmap: np.array):
|
48 |
fig, ax = plt.subplots()
|
49 |
cmap = plt.get_cmap("gist_heat")
|
50 |
+
h = -heatmap[0, 0].copy()
|
51 |
+
qmin, qmax = np.quantile(h, 0.8), np.quantile(h, 0.999)
|
52 |
h = np.clip(h, a_min=qmin, a_max=qmax)
|
53 |
+
h = (h - h.min()) / (h.max() - h.min())
|
54 |
+
h = cmap(h, bytes=True)[:, :, :3]
|
55 |
h = Image.fromarray(h).resize(img.size, resample=Image.Resampling.BILINEAR)
|
56 |
im = Image.blend(img, h, alpha=0.6)
|
57 |
+
# im = ax.imshow(np.array(im))
|
58 |
+
# # fig.colorbar(im)
|
59 |
+
# # plt.grid(False)
|
60 |
+
# # plt.axis("off")
|
61 |
+
# fig.tight_layout()
|
62 |
+
return im
|
63 |
|
64 |
+
|
65 |
+
def run_inference(input_img, preset="edm2-img64-s-fid", device="cuda"):
|
66 |
|
67 |
# img = center_crop_imagenet(64, img)
|
68 |
input_img = input_img.resize(size=(64, 64), resample=Image.Resampling.LANCZOS)
|
69 |
|
70 |
with torch.inference_mode():
|
71 |
img = np.array(input_img)
|
72 |
+
img = torch.from_numpy(img).permute(2, 0, 1).unsqueeze(0)
|
73 |
img = img.float().to(device)
|
74 |
+
model = load_model(modeldir="models", preset=preset, device=device)
|
75 |
img_likelihood = model(img).cpu().numpy()
|
76 |
+
# img_likelihood = model.scorenet(img).square().sum(1).sum(1).contiguous().float().cpu().unsqueeze(1).numpy()
|
77 |
+
# print(img_likelihood.shape, img_likelihood.dtype)
|
78 |
+
img = torch.nn.functional.interpolate(img, size=64, mode="bilinear")
|
79 |
x = model.scorenet(img)
|
80 |
x = x.square().sum(dim=(2, 3, 4)) ** 0.5
|
81 |
+
nll, pct, ref_nll = compute_gmm_likelihood(
|
82 |
+
x.cpu(), model_dir=f"models/{preset}"
|
83 |
+
)
|
84 |
|
85 |
outstr = f"Anomaly score: {nll:.3f} / {pct:.2f} percentile"
|
86 |
histplot = plot_against_reference(nll, ref_nll)
|
|
|
91 |
|
92 |
demo = gr.Interface(
|
93 |
fn=run_inference,
|
94 |
+
inputs=[
|
95 |
+
gr.Image(type="pil", label="Input Image"),
|
96 |
+
gr.Dropdown(choices=config_presets.keys(), label="Score Model"),
|
97 |
+
],
|
98 |
+
outputs=[
|
99 |
+
"text",
|
100 |
+
gr.Image(label="Anomaly Heatmap", min_width=64),
|
101 |
+
gr.Plot(label="Comparing to Imagenette"),
|
102 |
+
],
|
103 |
+
|
104 |
+
examples=[
|
105 |
+
['goldfish.JPEG', "edm2-img64-s-fid"]
|
106 |
+
]
|
107 |
)
|
108 |
|
109 |
if __name__ == "__main__":
|