Upload 4 files

S12_incorrect.csv

@@ -0,0 +1,742 @@
+indices,ground_truths,predicted_vals
+24,5,4
+35,2,3
+58,4,5
+59,6,3
+61,3,5
+86,2,7
+118,2,6
+143,3,5
+147,2,5
+158,5,3
+169,0,8
+178,5,2
+188,4,2
+226,6,0
+237,0,9
+275,5,3
+277,3,2
+308,1,9
+313,0,2
+376,0,9
+378,0,8
+384,2,5
+394,9,1
+422,0,8
+426,3,6
+428,0,6
+456,3,4
+470,3,5
+483,5,3
+497,2,5
+508,9,1
+531,0,3
+537,2,6
+551,5,3
+567,7,3
+577,9,8
+598,5,3
+618,2,3
+642,0,2
+655,2,4
+665,8,0
+672,6,2
+679,5,3
+683,8,0
+711,8,0
+725,2,0
+727,3,0
+731,2,0
+739,3,5
+751,2,0
+770,6,4
+771,1,8
+776,2,3
+779,5,6
+793,4,5
+799,2,4
+810,5,2
+811,3,2
+828,5,4
+845,3,2
+850,7,5
+863,3,5
+878,3,4
+882,3,0
+888,0,8
+893,6,3
+948,3,4
+972,2,6
+982,0,4
+997,1,9
+1040,7,5
+1042,5,2
+1057,7,5
+1066,9,1
+1072,0,8
+1074,3,6
+1088,3,5
+1090,0,8
+1095,2,6
+1100,7,5
+1118,4,7
+1126,2,0
+1129,3,5
+1131,1,9
+1137,0,9
+1147,6,9
+1163,3,5
+1177,9,1
+1224,7,4
+1227,3,5
+1230,5,3
+1265,9,1
+1292,4,3
+1304,4,8
+1325,2,4
+1334,3,2
+1346,3,5
+1391,7,5
+1393,4,2
+1394,6,5
+1416,3,5
+1442,3,5
+1445,3,4
+1461,0,9
+1470,7,3
+1492,9,1
+1494,7,3
+1495,2,0
+1501,2,6
+1503,4,3
+1506,5,7
+1536,5,3
+1552,3,5
+1580,6,2
+1620,5,8
+1628,4,7
+1644,2,3
+1684,1,8
+1685,8,2
+1718,0,8
+1723,2,5
+1727,9,1
+1731,5,3
+1732,9,1
+1734,9,1
+1741,4,5
+1746,2,0
+1779,3,5
+1803,2,3
+1804,5,0
+1812,5,3
+1823,0,2
+1830,3,5
+1845,7,4
+1858,4,7
+1862,5,4
+1884,7,5
+1885,5,3
+1886,8,2
+1895,8,2
+1897,8,0
+1909,3,4
+1935,6,2
+1938,8,0
+1939,6,7
+1945,4,7
+1957,3,5
+1969,1,9
+1983,8,2
+2001,5,4
+2010,7,2
+2034,3,4
+2036,5,3
+2059,9,8
+2061,5,3
+2067,7,4
+2078,2,6
+2111,9,1
+2128,4,6
+2157,3,5
+2161,5,3
+2165,2,0
+2172,5,2
+2175,3,5
+2187,5,3
+2216,2,3
+2226,2,6
+2232,0,3
+2248,0,8
+2251,6,3
+2253,3,5
+2270,2,4
+2271,9,0
+2293,8,1
+2299,6,3
+2300,3,0
+2350,3,0
+2362,3,4
+2363,2,6
+2379,0,8
+2386,4,3
+2405,3,6
+2409,4,2
+2413,4,6
+2442,4,3
+2447,6,2
+2472,5,3
+2473,0,8
+2495,9,8
+2511,8,1
+2525,5,2
+2532,0,1
+2533,8,9
+2535,0,8
+2542,2,0
+2580,6,3
+2586,7,3
+2590,5,0
+2592,2,7
+2597,2,3
+2634,1,9
+2650,5,0
+2675,2,6
+2705,8,0
+2719,4,3
+2729,5,3
+2770,7,3
+2779,3,5
+2785,4,5
+2798,3,5
+2804,3,5
+2831,3,0
+2843,1,8
+2844,4,3
+2845,5,2
+2854,9,1
+2895,7,5
+2912,2,6
+2923,3,5
+2937,3,5
+2940,3,0
+2948,5,4
+2958,0,8
+2960,9,0
+3013,3,5
+3019,5,3
+3025,2,6
+3026,5,7
+3032,2,6
+3049,3,5
+3051,8,2
+3052,3,6
+3059,5,3
+3068,2,0
+3075,6,3
+3083,5,3
+3094,2,0
+3095,4,6
+3107,5,3
+3115,5,3
+3150,1,9
+3158,0,8
+3180,3,5
+3183,3,5
+3192,5,4
+3202,5,3
+3231,3,5
+3236,5,3
+3301,8,2
+3342,0,8
+3354,3,5
+3359,7,3
+3390,3,5
+3400,6,0
+3412,0,3
+3416,7,4
+3439,4,6
+3443,4,2
+3444,0,2
+3450,9,1
+3459,9,8
+3496,5,3
+3514,2,7
+3524,5,4
+3550,0,4
+3560,1,9
+3574,5,3
+3584,3,5
+3594,3,8
+3596,2,4
+3600,4,3
+3601,0,7
+3607,3,2
+3615,6,0
+3653,8,0
+3658,2,4
+3669,5,3
+3696,4,3
+3704,5,4
+3708,4,3
+3716,3,6
+3726,5,3
+3728,2,4
+3735,5,3
+3753,4,6
+3774,5,3
+3779,3,0
+3785,6,2
+3789,9,1
+3799,3,5
+3801,2,0
+3821,8,3
+3828,1,0
+3887,2,4
+3899,6,2
+3957,9,0
+3977,3,6
+3980,9,0
+3995,7,3
+3996,6,2
+4000,8,7
+4003,2,0
+4012,3,5
+4013,1,9
+4036,7,0
+4051,5,3
+4054,7,0
+4055,1,0
+4071,3,2
+4101,2,0
+4106,3,4
+4112,4,5
+4125,8,0
+4135,1,0
+4137,4,3
+4139,5,0
+4140,5,3
+4152,5,3
+4169,3,5
+4190,3,5
+4244,2,7
+4251,8,0
+4255,9,6
+4276,9,1
+4302,2,5
+4309,7,5
+4315,6,0
+4404,3,4
+4409,8,1
+4423,6,2
+4446,0,8
+4463,5,2
+4483,2,6
+4485,3,7
+4526,3,5
+4545,3,4
+4546,2,0
+4574,3,4
+4590,9,0
+4599,8,1
+4606,0,3
+4614,3,5
+4618,3,6
+4628,5,7
+4630,2,6
+4652,3,5
+4669,6,4
+4710,3,7
+4729,2,3
+4740,5,3
+4744,6,0
+4748,0,3
+4749,2,3
+4754,3,8
+4760,3,0
+4776,3,5
+4784,0,8
+4786,6,3
+4794,4,2
+4795,5,3
+4862,3,0
+4887,5,3
+4892,7,4
+4906,5,3
+4917,5,3
+4921,5,6
+4931,9,1
+4942,5,8
+4964,5,4
+4965,3,4
+4970,2,3
+4995,2,3
+4996,3,5
+5006,7,2
+5068,4,5
+5098,5,2
+5117,7,3
+5119,5,0
+5134,5,3
+5151,9,3
+5162,7,0
+5176,3,9
+5183,4,3
+5187,4,5
+5193,4,2
+5206,3,5
+5211,7,4
+5218,3,5
+5224,0,3
+5229,5,3
+5238,2,0
+5279,0,9
+5290,3,7
+5292,4,7
+5324,2,4
+5330,7,5
+5340,3,4
+5369,4,6
+5388,6,3
+5398,8,0
+5410,3,5
+5416,9,1
+5417,4,3
+5426,5,3
+5433,3,5
+5454,0,2
+5455,9,1
+5458,3,6
+5468,5,3
+5511,3,6
+5537,2,3
+5548,8,0
+5565,3,6
+5582,4,3
+5603,5,4
+5607,6,0
+5609,5,3
+5628,8,1
+5632,3,5
+5642,5,4
+5659,2,6
+5708,2,5
+5709,2,4
+5718,1,8
+5730,5,3
+5745,5,3
+5747,8,0
+5749,5,3
+5794,7,0
+5808,5,3
+5830,5,3
+5835,3,5
+5856,3,5
+5859,2,3
+5862,2,3
+5870,7,4
+5881,3,5
+5886,2,4
+5908,3,6
+5910,2,5
+5915,5,3
+5924,4,2
+5927,6,2
+5960,2,6
+5987,3,7
+5991,3,5
+5992,4,7
+6008,3,0
+6014,5,4
+6025,6,3
+6028,8,2
+6051,4,7
+6063,6,4
+6068,5,3
+6070,8,9
+6083,2,3
+6127,5,2
+6144,7,5
+6145,2,5
+6146,0,9
+6151,7,6
+6160,2,0
+6162,0,9
+6174,3,5
+6180,0,8
+6202,3,2
+6210,4,6
+6243,4,6
+6257,5,2
+6283,3,5
+6305,6,3
+6382,4,3
+6396,4,6
+6401,4,7
+6423,5,4
+6439,3,4
+6446,4,2
+6466,8,0
+6480,0,8
+6498,4,6
+6499,7,3
+6522,9,1
+6545,9,8
+6559,2,4
+6562,5,3
+6588,0,8
+6596,3,2
+6653,0,2
+6669,5,4
+6709,3,5
+6729,0,7
+6741,9,0
+6748,0,6
+6750,4,3
+6753,4,2
+6759,1,9
+6783,3,2
+6792,3,9
+6859,3,1
+6861,8,0
+6876,7,9
+6901,2,9
+6912,3,5
+6942,3,2
+6962,7,4
+6969,2,6
+6970,0,2
+6973,3,5
+6979,5,4
+6980,0,8
+6994,3,4
+7000,2,8
+7002,3,4
+7004,3,7
+7005,1,9
+7035,1,9
+7037,7,5
+7040,7,0
+7068,2,0
+7090,2,5
+7099,3,5
+7107,2,6
+7112,2,0
+7143,4,2
+7150,3,6
+7168,5,3
+7214,3,5
+7217,8,6
+7228,3,2
+7265,5,7
+7274,1,9
+7311,1,9
+7367,3,5
+7384,3,5
+7395,2,3
+7398,5,2
+7399,3,5
+7400,3,5
+7408,4,5
+7411,5,3
+7414,8,0
+7420,3,5
+7444,9,1
+7496,3,4
+7509,3,5
+7519,2,8
+7524,3,8
+7545,5,3
+7556,2,0
+7559,8,1
+7561,2,3
+7562,2,3
+7590,2,4
+7600,8,1
+7605,2,6
+7617,2,0
+7620,5,3
+7644,3,5
+7657,2,5
+7685,9,0
+7705,7,5
+7711,6,2
+7744,4,8
+7760,2,0
+7761,2,3
+7763,4,2
+7784,7,2
+7792,9,1
+7797,3,6
+7815,0,2
+7816,0,4
+7827,3,5
+7839,3,4
+7852,5,7
+7853,3,2
+7881,3,4
+7892,1,9
+7924,3,5
+7941,5,2
+7947,2,3
+7958,5,3
+7967,9,1
+7983,6,8
+8014,2,4
+8030,2,0
+8033,3,5
+8046,8,9
+8058,4,7
+8084,5,3
+8115,5,3
+8119,6,2
+8131,3,6
+8166,2,6
+8198,4,0
+8209,2,4
+8236,5,2
+8245,3,5
+8254,8,0
+8266,5,3
+8269,8,0
+8281,3,5
+8295,3,0
+8300,5,3
+8305,4,3
+8314,6,3
+8317,5,3
+8324,3,5
+8362,3,5
+8363,2,6
+8420,3,6
+8428,5,6
+8433,3,6
+8451,5,7
+8457,2,4
+8473,2,4
+8480,2,6
+8490,3,5
+8507,8,0
+8514,7,3
+8521,9,8
+8529,0,9
+8536,1,9
+8542,2,6
+8548,3,4
+8573,6,0
+8576,2,0
+8577,3,6
+8578,2,5
+8598,3,2
+8629,7,4
+8633,3,8
+8644,8,0
+8650,2,5
+8671,4,6
+8699,4,5
+8723,5,4
+8727,9,0
+8728,6,5
+8735,3,2
+8743,0,8
+8757,8,0
+8773,6,2
+8778,3,0
+8779,7,5
+8803,3,5
+8827,2,5
+8840,2,3
+8847,5,3
+8848,3,7
+8854,4,8
+8858,6,4
+8898,7,0
+8901,5,7
+8928,4,2
+8930,3,8
+8932,6,2
+8941,0,3
+8943,7,5
+8958,6,2
+8976,7,2
+8977,0,8
+8983,3,4
+8985,9,0
+8993,0,9
+9026,9,0
+9031,0,8
+9034,2,4
+9039,2,6
+9067,9,1
+9080,3,5
+9082,2,3
+9090,2,4
+9105,8,0
+9107,5,4
+9132,2,6
+9145,1,9
+9148,3,5
+9157,3,5
+9190,3,4
+9209,9,1
+9216,6,3
+9218,3,5
+9223,1,9
+9225,5,8
+9227,1,9
+9231,7,2
+9254,7,5
+9260,5,3
+9294,6,3
+9305,5,7
+9319,3,5
+9336,3,5
+9360,5,3
+9369,5,4
+9375,3,5
+9385,7,3
+9386,4,2
+9414,5,3
+9431,3,5
+9497,3,6
+9501,6,0
+9503,2,4
+9505,3,5
+9506,4,2
+9518,9,1
+9522,3,2
+9535,1,9
+9563,1,5
+9587,7,5
+9616,5,3
+9625,2,0
+9633,0,3
+9641,6,3
+9643,9,0
+9657,5,3
+9665,3,5
+9704,2,3
+9707,5,7
+9729,4,6
+9734,4,7
+9750,2,3
+9753,3,2
+9786,6,1
+9787,8,0
+9805,2,0
+9812,3,8
+9817,9,1
+9819,3,5
+9832,2,5
+9840,4,7
+9842,6,4
+9853,5,6
+9857,0,1
+9884,9,1
+9901,3,6
+9910,4,7
+9949,3,2
+9960,2,0
+9968,3,5
+9982,2,6
+9989,2,4

app.py

@@ -0,0 +1,94 @@
+import torch
+import pandas as pd
+import numpy as np
+import gradio as gr
+from PIL import Image
+from torch.nn import functional as F
+from collections import OrderedDict
+from torchvision import transforms
+from pytorch_grad_cam import GradCAM
+from pytorch_grad_cam.utils.image import show_cam_on_image
+from pytorch_grad_cam.utils.model_targets import ClassifierOutputTarget
+
+from models.custom_resnet import Model
+from datasets import CIFAR10
+
+cifar10 = CIFAR10(normalize=False, shuffle=False, augment=False)
+_ = cifar10.test_data
+
+missed_df = pd.read_csv('S12_incorrect.csv')
+missed_df['ground_truths'] = missed_df['ground_truths'].map(cifar10.classes)
+missed_df['predicted_vals'] = missed_df['predicted_vals'].map(cifar10.classes)
+missed_df = missed_df.sample(frac=1)
+
+model = Model(cifar10)
+model.load_state_dict(torch.load('S12_model.pth', map_location='mps'))
+model.eval()
+
+transform = transforms.Compose([
+    transforms.ToTensor(),
+    transforms.Normalize(mean=[0.5, 0.5, 0.5], std=[0.25, 0.25, 0.25])
+])
+
+inv_transform = transforms.Normalize(mean=[-2, -2, -2], std=[4, 4, 4])
+
+
+def image_classifier(input_image, top_classes=3, show_cam=True, target_layer=-1, transparency=0.5):
+    input = transform(input_image).unsqueeze(0)
+
+    output = model(input)
+    output = F.softmax(output.flatten(), dim=-1)
+
+    confidences = [(cifar10.classes[i], float(output[i])) for i in range(10)]
+    confidences.sort(key=lambda x: x[1], reverse=True)
+    confidences = OrderedDict(confidences[:top_classes])
+
+    label = torch.argmax(output).item()
+    target_layer = [model.network[4 + target_layer]]
+    grad_cam = GradCAM(model=model, target_layers=target_layer, use_cuda=False)
+    targets = [ClassifierOutputTarget(label)]
+    grayscale_cam = grad_cam(input_tensor=input, targets=targets)
+    grayscale_cam = grayscale_cam[0, :]
+    output_image = show_cam_on_image(input_image/255, grayscale_cam, use_rgb=True, image_weight=transparency)
+
+    return output_image if show_cam else input_image, confidences
+
+
+demo1 = gr.Interface(
+    fn=image_classifier,
+    inputs=[
+        gr.Image(shape=(32, 32), label="Input Image", value='examples/cat.jpg'),
+        gr.Slider(1, 10, value = 3, step=1, label="Number of Top Classes"),
+        gr.Checkbox(label="Show GradCAM?", value=True),
+        gr.Slider(-4, -1, value = -1, step=1, label="Which Layer?"),
+        gr.Slider(0, 1, value = 0.7, label="Transparncy", step=0.1)
+    ],
+    outputs=[gr.Image(shape=(32, 32), label="Output Image"),
+             gr.Label(label='Top Classes')],
+    examples=[[f'examples/{k}.jpg'] for k in cifar10.classes.values()]
+)
+
+
+def show_incorrect(num_examples=10, show_cam=True, target_layer=-1, transparency=0.5):
+    result = list()
+    for index, row in missed_df.head(num_examples).iterrows():
+        image = np.asarray(Image.open(f'missed_examples/{index}.jpg'))
+        output_image, confidence = image_classifier(image, show_cam=show_cam, target_layer=target_layer, transparency=transparency)
+        predicted = list(confidence)[0]
+        result.append((output_image, f"{row['ground_truths']} / {predicted}"))
+    return result
+
+
+demo2 = gr.Interface(
+    fn=show_incorrect,
+    inputs=[
+        gr.Number(value=10, minimum=1, maximum=len(missed_df), label="No. of missclassified Examples", precision=0),
+        gr.Checkbox(label="Show GradCAM?", value=True),
+        gr.Slider(-4, -1, value = -1, step=1, label="Which Layer?"),
+        gr.Slider(0, 1, value = 0.7, label="Transparncy", step=0.1),
+    ],
+    outputs=[gr.Gallery(label="Missclassified Images (Truth / Predicted)", columns=4)]
+)
+
+demo = gr.TabbedInterface([demo1, demo2], ["Examples", "Misclassified Examples"])
+demo.launch()

requirements.txt

@@ -0,0 +1,15 @@
+torch
+torchvision
+torchinfo
+tqdm
+matplotlib
+albumentations
+numpy
+opencv-python
+torch-lr-finder
+grad-cam
+pytorch-lightning
+torchmetrics
+pandas
+gradio
+Pillow