-
Notifications
You must be signed in to change notification settings - Fork 30
/
README.html
50 lines (47 loc) · 2.85 KB
/
README.html
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>README</title>
<link rel="stylesheet" href="https://stackedit.io/style.css" />
</head>
<body class="stackedit">
<div class="stackedit__html"><h1 id="advanced-training">Advanced Training</h1>
<p>Assignment:</p>
<ol>
<li>Pick your last code</li>
<li>Make sure to Add CutOut to your code. It should come from your transformations (albumentations)</li>
<li>Use this repo: <a href="https://github.com/davidtvs/pytorch-lr-finder">https://github.com/davidtvs/pytorch-lr-finder (Links to an external site.)</a>
<ol>
<li>Move LR Finder code to your modules</li>
<li>Implement LR Finder (for SGD, not for ADAM)</li>
<li>Implement ReduceLROnPlatea: <a href="https://pytorch.org/docs/stable/optim.html#torch.optim.lr_scheduler.ReduceLROnPlateau">https://pytorch.org/docs/stable/optim.html#torch.optim.lr_scheduler.ReduceLROnPlateau (Links to an external site.)</a></li>
</ol>
</li>
<li>Find best LR to train your model</li>
<li>Use SDG with Momentum</li>
<li>Train for 50 Epochs.</li>
<li>Show Training and Test Accuracy curves</li>
<li>Target 88% Accuracy.</li>
<li>Run GradCAM on the any 25 misclassified images. Make sure you mention what is the prediction and what was the ground truth label.</li>
<li>Submit</li>
</ol>
<h1 id="solution">Solution</h1>
<p>Github-Notebook : <a href="https://github.com/satyajitghana/TSAI-DeepVision-EVA4.0/blob/master/10_AdvancedTraining/AdvancedTraining.ipynb">https://github.com/satyajitghana/TSAI-DeepVision-EVA4.0/blob/master/10_AdvancedTraining/AdvancedTraining.ipynb</a></p>
<p>Google-Colab: <a href="https://colab.research.google.com/github/satyajitghana/TSAI-DeepVision-EVA4.0/blob/master/10_AdvancedTraining/AdvancedTraining.ipynb">https://colab.research.google.com/github/satyajitghana/TSAI-DeepVision-EVA4.0/blob/master/10_AdvancedTraining/AdvancedTraining.ipynb</a></p>
<h2 id="model-stats">Model Stats</h2>
<pre><code>Epochs: 50
Max Train Accuracy: 98.27
Max Test Accuracy: 92.03
</code></pre>
<h2 id="model-metrics">Model Metrics</h2>
<p><img src="https://github.com/satyajitghana/TSAI-DeepVision-EVA4.0/blob/master/10_AdvancedTraining/metrics_1.png?raw=true" alt="enter image description here"></p>
<p><img src="https://github.com/satyajitghana/TSAI-DeepVision-EVA4.0/blob/master/10_AdvancedTraining/metrics_2.png?raw=true" alt="enter image description here"></p>
<h2 id="grad-cam">Grad-CAM</h2>
<p><img src="https://github.com/satyajitghana/TSAI-DeepVision-EVA4.0/blob/master/10_AdvancedTraining/gradcam.png?raw=true" alt="enter image description here"></p>
<h2 id="misclassifications-grad-cam">Misclassifications Grad-CAM</h2>
<p><img src="https://github.com/satyajitghana/TSAI-DeepVision-EVA4.0/blob/master/10_AdvancedTraining/misclassified_gradcam.png?raw=true" alt="enter image description here"></p>
</div>
</body>
</html>