Spaces:
Sleeping
Sleeping
Commit
·
74a7e82
1
Parent(s):
e67bbc1
init demo
Browse files- app.py +174 -0
- images/examples/cat.jpg +0 -0
- images/examples/dog.jpg +0 -0
- images/miss_classified/airplane_0.png +0 -0
- images/miss_classified/airplane_6.png +0 -0
- images/miss_classified/bird_12.png +0 -0
- images/miss_classified/bird_19.png +0 -0
- images/miss_classified/bird_2.png +0 -0
- images/miss_classified/car_1.png +0 -0
- images/miss_classified/cat_10.png +0 -0
- images/miss_classified/cat_14.png +0 -0
- images/miss_classified/cat_17.png +0 -0
- images/miss_classified/cat_5.png +0 -0
- images/miss_classified/deer_15.png +0 -0
- images/miss_classified/deer_8.png +0 -0
- images/miss_classified/dog_11.png +0 -0
- images/miss_classified/dog_13.png +0 -0
- images/miss_classified/dog_18.png +0 -0
- images/miss_classified/dog_4.png +0 -0
- images/miss_classified/horse_16.png +0 -0
- images/miss_classified/horse_3.png +0 -0
- images/miss_classified/truck_7.png +0 -0
- images/miss_classified/truck_9.png +0 -0
- mini_resnet.py +89 -0
- weights/weights.pt +3 -0
app.py
ADDED
|
@@ -0,0 +1,174 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import os
|
| 2 |
+
from io import BytesIO
|
| 3 |
+
from pathlib import Path
|
| 4 |
+
from random import shuffle
|
| 5 |
+
|
| 6 |
+
import cv2
|
| 7 |
+
import gradio as gr
|
| 8 |
+
import matplotlib.pyplot as plt
|
| 9 |
+
import numpy as np
|
| 10 |
+
import torch
|
| 11 |
+
from mini_resnet import CustomResNet
|
| 12 |
+
from PIL import Image
|
| 13 |
+
from pytorch_grad_cam import GradCAM
|
| 14 |
+
from pytorch_grad_cam.utils.image import show_cam_on_image
|
| 15 |
+
from torchvision import transforms as T
|
| 16 |
+
|
| 17 |
+
mean = (0.49139968, 0.48215841, 0.44653091)
|
| 18 |
+
std = (0.24703223, 0.24348513, 0.26158784)
|
| 19 |
+
transforms = T.Compose([T.ToTensor(), T.Normalize(mean=mean, std=std)])
|
| 20 |
+
classes = ("plane", "car", "bird", "cat", "deer", "dog", "frog", "horse", "ship", "truck")
|
| 21 |
+
softmax = torch.nn.Softmax(dim=0)
|
| 22 |
+
|
| 23 |
+
model = CustomResNet()
|
| 24 |
+
model.load_state_dict(torch.load("weights/weights.pt", map_location=torch.device("cpu")))
|
| 25 |
+
model.eval()
|
| 26 |
+
|
| 27 |
+
misclf_path = "images/miss_classified"
|
| 28 |
+
mis_classified_imgs = list(Path(misclf_path).glob("*"))
|
| 29 |
+
|
| 30 |
+
|
| 31 |
+
def get_traget_layer(block: str, layer: int):
|
| 32 |
+
layer_num = 0 if layer == 0 else -1
|
| 33 |
+
if block == "block1":
|
| 34 |
+
return model.layer1[layer_num]
|
| 35 |
+
if block == "block2":
|
| 36 |
+
return model.layer2[layer_num]
|
| 37 |
+
if block == "block3":
|
| 38 |
+
return model.layer3[layer_num]
|
| 39 |
+
|
| 40 |
+
|
| 41 |
+
default_cam = GradCAM(model=model, target_layers=[get_traget_layer("block3", -1)])
|
| 42 |
+
|
| 43 |
+
|
| 44 |
+
def make_image(p: Path | str, pred: str, label: str):
|
| 45 |
+
im = cv2.imread(str(p))
|
| 46 |
+
im = cv2.resize(im, (64, 64))
|
| 47 |
+
|
| 48 |
+
plt.imshow(im)
|
| 49 |
+
plt.title(f"{pred} / {label}")
|
| 50 |
+
plt.axis("off")
|
| 51 |
+
|
| 52 |
+
buffer = BytesIO()
|
| 53 |
+
plt.savefig(buffer, format="png")
|
| 54 |
+
buffer.seek(0)
|
| 55 |
+
|
| 56 |
+
img_array = np.frombuffer(buffer.getvalue(), dtype=np.uint8)
|
| 57 |
+
buffer.close()
|
| 58 |
+
|
| 59 |
+
# Decode the image array using OpenCV
|
| 60 |
+
im = cv2.imdecode(img_array, cv2.IMREAD_COLOR)
|
| 61 |
+
return im
|
| 62 |
+
|
| 63 |
+
|
| 64 |
+
@torch.inference_mode()
|
| 65 |
+
def predict_img(img: np.ndarray, top_k: int = 10):
|
| 66 |
+
preds = model(img)
|
| 67 |
+
preds = softmax(preds.flatten())
|
| 68 |
+
preds = {classes[i]: float(preds[i]) for i in range(10)}
|
| 69 |
+
preds = {
|
| 70 |
+
k: v for k, v in sorted(preds.items(), key=lambda item: item[1], reverse=True)[:top_k]
|
| 71 |
+
}
|
| 72 |
+
|
| 73 |
+
return preds
|
| 74 |
+
|
| 75 |
+
|
| 76 |
+
def display_cam(cam: GradCAM, org_img: np.ndarray, img: torch.Tensor, transparency: float):
|
| 77 |
+
grayscale_cam = cam(input_tensor=img, targets=None)
|
| 78 |
+
grayscale_cam = grayscale_cam[0, :]
|
| 79 |
+
visualization = show_cam_on_image(
|
| 80 |
+
org_img / 255, grayscale_cam, use_rgb=True, image_weight=transparency
|
| 81 |
+
)
|
| 82 |
+
return visualization
|
| 83 |
+
|
| 84 |
+
|
| 85 |
+
def inference(
|
| 86 |
+
org_img: np.ndarray,
|
| 87 |
+
top_k: int,
|
| 88 |
+
show_cam: str,
|
| 89 |
+
num_cam_imgs: int,
|
| 90 |
+
cam_block: str,
|
| 91 |
+
target_layer_num: int,
|
| 92 |
+
transparency: float,
|
| 93 |
+
show_misclf: str,
|
| 94 |
+
num_misclf: int,
|
| 95 |
+
):
|
| 96 |
+
input_img = transforms(org_img)
|
| 97 |
+
input_img = input_img.unsqueeze(0)
|
| 98 |
+
|
| 99 |
+
preds = predict_img(input_img, top_k)
|
| 100 |
+
org_img = display_cam(default_cam, org_img, input_img, transparency)
|
| 101 |
+
|
| 102 |
+
shuffle(mis_classified_imgs)
|
| 103 |
+
cam_outputs = []
|
| 104 |
+
if show_cam:
|
| 105 |
+
img_list = []
|
| 106 |
+
|
| 107 |
+
target_layers = [get_traget_layer(cam_block, target_layer_num)]
|
| 108 |
+
cam = GradCAM(model=model, target_layers=target_layers, use_cuda=False)
|
| 109 |
+
for p in mis_classified_imgs[:num_cam_imgs]:
|
| 110 |
+
im = cv2.imread(str(p))
|
| 111 |
+
inp_im = transforms(im)
|
| 112 |
+
inp_im = inp_im.unsqueeze(0)
|
| 113 |
+
|
| 114 |
+
grayscale_cam = cam(input_tensor=inp_im, targets=None)
|
| 115 |
+
|
| 116 |
+
grayscale_cam = grayscale_cam[0, :]
|
| 117 |
+
visualization = show_cam_on_image(
|
| 118 |
+
im / 255, grayscale_cam, use_rgb=True, image_weight=transparency
|
| 119 |
+
)
|
| 120 |
+
cam_outputs.append(visualization)
|
| 121 |
+
|
| 122 |
+
del cam, img_list
|
| 123 |
+
|
| 124 |
+
misclf_images_output = []
|
| 125 |
+
if show_misclf:
|
| 126 |
+
img_list = []
|
| 127 |
+
gt = []
|
| 128 |
+
for p in mis_classified_imgs[:num_misclf]:
|
| 129 |
+
img_list.append(transforms(Image.open(p).convert("RGB")))
|
| 130 |
+
gt.append(p.name.split("_")[0])
|
| 131 |
+
|
| 132 |
+
misclf_out = softmax(model(torch.stack(img_list))).argmax(dim=1).tolist()
|
| 133 |
+
del img_list
|
| 134 |
+
for imp, pred, label in zip(mis_classified_imgs[:num_misclf], misclf_out, gt):
|
| 135 |
+
pred = classes[pred]
|
| 136 |
+
misclf_images_output.append(make_image(imp, pred, label))
|
| 137 |
+
|
| 138 |
+
return org_img, preds, cam_outputs, misclf_images_output
|
| 139 |
+
|
| 140 |
+
|
| 141 |
+
title = "CIFAR10 trained on Custom Model inspired by ResNet with GradCAM"
|
| 142 |
+
description = "A simple Gradio interface to infer on ResNet model, and get GradCAM results. You can see the code <a href='https://github.com/anantgupta129/TorcHood'>here</a> & <a href='https://colab.research.google.com/github/anantgupta129/ERA-V1/blob/main/session12/notebooks/s12_train.ipynb'>training notebook</a>"
|
| 143 |
+
|
| 144 |
+
examples = [["images/examples/cat.jpg", 3, True, 5, "block3", 1, 0.5, True, 5], ["images/examples/dog.jpg", 5, True, 5, "block3", 1, 0.5, True, 5]]
|
| 145 |
+
demo = gr.Interface(
|
| 146 |
+
inference,
|
| 147 |
+
inputs=[
|
| 148 |
+
gr.Image(shape=(32, 32), label="Input Image"),
|
| 149 |
+
gr.Slider(1, 10, value=3, step=1, label="Top K predictions"),
|
| 150 |
+
gr.Checkbox(label="Show Grad Cam"),
|
| 151 |
+
gr.Slider(1, 20, value=5, step=1, label="Number of images"),
|
| 152 |
+
gr.Radio(label="Which Block?", choices=["block1", "block2", "block3"]),
|
| 153 |
+
gr.Slider(0, 1, value=1, step=1, label="Which Layer?"),
|
| 154 |
+
gr.Slider(0, 1, value=0.5, label="Opacity of GradCAM"),
|
| 155 |
+
gr.Checkbox(label="Show Misclassified Images"),
|
| 156 |
+
gr.Slider(1, 20, value=5, step=5, label="Number of Misclassification Images"),
|
| 157 |
+
],
|
| 158 |
+
outputs=[
|
| 159 |
+
gr.Image(shape=(32, 32), label="Output", width=128, height=128),
|
| 160 |
+
"label",
|
| 161 |
+
gr.Gallery(label="GradCAM Output"),
|
| 162 |
+
gr.Gallery(
|
| 163 |
+
label="Misclassified Images Pred/G.T.",
|
| 164 |
+
columns=[2],
|
| 165 |
+
rows=[2],
|
| 166 |
+
object_fit="contain",
|
| 167 |
+
height="auto",
|
| 168 |
+
),
|
| 169 |
+
],
|
| 170 |
+
title=title,
|
| 171 |
+
description=description,
|
| 172 |
+
examples=examples,
|
| 173 |
+
)
|
| 174 |
+
demo.launch()
|
images/examples/cat.jpg
ADDED
|
images/examples/dog.jpg
ADDED
|
images/miss_classified/airplane_0.png
ADDED
|
images/miss_classified/airplane_6.png
ADDED
|
images/miss_classified/bird_12.png
ADDED
|
images/miss_classified/bird_19.png
ADDED
|
images/miss_classified/bird_2.png
ADDED
|
images/miss_classified/car_1.png
ADDED
|
images/miss_classified/cat_10.png
ADDED
|
images/miss_classified/cat_14.png
ADDED
|
images/miss_classified/cat_17.png
ADDED
|
images/miss_classified/cat_5.png
ADDED
|
images/miss_classified/deer_15.png
ADDED
|
images/miss_classified/deer_8.png
ADDED
|
images/miss_classified/dog_11.png
ADDED
|
images/miss_classified/dog_13.png
ADDED
|
images/miss_classified/dog_18.png
ADDED
|
images/miss_classified/dog_4.png
ADDED
|
images/miss_classified/horse_16.png
ADDED
|
images/miss_classified/horse_3.png
ADDED
|
images/miss_classified/truck_7.png
ADDED
|
images/miss_classified/truck_9.png
ADDED
|
mini_resnet.py
ADDED
|
@@ -0,0 +1,89 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import torch
|
| 2 |
+
import torch.nn as nn
|
| 3 |
+
import torch.nn.functional as F
|
| 4 |
+
|
| 5 |
+
# from common import BaseNet
|
| 6 |
+
|
| 7 |
+
|
| 8 |
+
class ResBlock(nn.Module):
|
| 9 |
+
def __init__(self, in_planes: int, out_planes: int, stride: int = 1, drop: float = 0) -> None:
|
| 10 |
+
super().__init__()
|
| 11 |
+
self.dropout = nn.Dropout2d(drop)
|
| 12 |
+
|
| 13 |
+
self.conv1 = nn.Conv2d(
|
| 14 |
+
in_planes,
|
| 15 |
+
out_planes,
|
| 16 |
+
kernel_size=3,
|
| 17 |
+
stride=stride,
|
| 18 |
+
padding=1,
|
| 19 |
+
bias=False,
|
| 20 |
+
)
|
| 21 |
+
self.bn1 = nn.BatchNorm2d(out_planes)
|
| 22 |
+
|
| 23 |
+
self.conv2 = nn.Conv2d(
|
| 24 |
+
out_planes,
|
| 25 |
+
out_planes,
|
| 26 |
+
kernel_size=3,
|
| 27 |
+
stride=stride,
|
| 28 |
+
padding=1,
|
| 29 |
+
bias=False,
|
| 30 |
+
)
|
| 31 |
+
self.bn2 = nn.BatchNorm2d(out_planes)
|
| 32 |
+
|
| 33 |
+
def forward(self, x: torch.Tensor) -> torch.Tensor:
|
| 34 |
+
out = F.relu(self.bn1(self.conv1(x)))
|
| 35 |
+
out = self.dropout(out)
|
| 36 |
+
out = self.bn2(self.conv2(out))
|
| 37 |
+
out += x
|
| 38 |
+
out = F.relu(out)
|
| 39 |
+
out = self.dropout(out)
|
| 40 |
+
|
| 41 |
+
return out
|
| 42 |
+
|
| 43 |
+
|
| 44 |
+
class CustomResNet(nn.Module):
|
| 45 |
+
def __init__(self, drop: float = 0, num_classes: int = 10) -> None:
|
| 46 |
+
super().__init__()
|
| 47 |
+
|
| 48 |
+
# perp layer
|
| 49 |
+
self.perlayer = nn.Sequential(
|
| 50 |
+
nn.Conv2d(3, 64, 3, padding=1, bias=False),
|
| 51 |
+
nn.BatchNorm2d(64),
|
| 52 |
+
nn.ReLU(),
|
| 53 |
+
nn.Dropout2d(drop),
|
| 54 |
+
)
|
| 55 |
+
self.layer1 = nn.Sequential(
|
| 56 |
+
nn.Conv2d(64, 128, 3, padding=1, bias=False),
|
| 57 |
+
nn.MaxPool2d(2, 2),
|
| 58 |
+
nn.BatchNorm2d(128),
|
| 59 |
+
nn.ReLU(),
|
| 60 |
+
nn.Dropout2d(drop),
|
| 61 |
+
ResBlock(128, 128, drop=drop),
|
| 62 |
+
)
|
| 63 |
+
self.layer2 = nn.Sequential(
|
| 64 |
+
nn.Conv2d(128, 256, 3, padding=1, bias=False),
|
| 65 |
+
nn.MaxPool2d(2, 2),
|
| 66 |
+
nn.BatchNorm2d(256),
|
| 67 |
+
nn.ReLU(),
|
| 68 |
+
nn.Dropout2d(drop),
|
| 69 |
+
)
|
| 70 |
+
self.layer3 = nn.Sequential(
|
| 71 |
+
nn.Conv2d(256, 512, 3, padding=1, bias=False),
|
| 72 |
+
nn.MaxPool2d(2, 2),
|
| 73 |
+
nn.BatchNorm2d(512),
|
| 74 |
+
nn.ReLU(),
|
| 75 |
+
nn.Dropout2d(drop),
|
| 76 |
+
ResBlock(512, 512, drop=drop),
|
| 77 |
+
)
|
| 78 |
+
self.pool = nn.MaxPool2d(4)
|
| 79 |
+
self.out = nn.Conv2d(512, num_classes, 1, bias=False)
|
| 80 |
+
|
| 81 |
+
def forward(self, x: torch.Tensor) -> torch.Tensor:
|
| 82 |
+
x = self.perlayer(x)
|
| 83 |
+
x = self.layer1(x)
|
| 84 |
+
x = self.layer2(x)
|
| 85 |
+
x = self.layer3(x)
|
| 86 |
+
x = self.pool(x)
|
| 87 |
+
x = self.out(x)
|
| 88 |
+
|
| 89 |
+
return x.view(-1, 10)
|
weights/weights.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:537219db09391c540e4a68ace7708857fb1dce2f5b0e0d325eba082bafc2643e
|
| 3 |
+
size 26325330
|