Skip to content
Project code for Stanford's CS231N Convnets for Visual Recognition (Spring 2017)
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
assignment1
assignment2
assignment3
project
.gitignore
README.md

README.md

cs231n-spring-2017

Project code for Stanford's CS231N Convnets for Visual Recognition (Spring 2017)

HW1 Score (out of 100): 100
HW1 Bonus (out of 10, if applicable): 0

Q1: k-Nearest Neighbor (kNN) classifier 20 / 20.0 pts
Q2: Support Vector Machine (SVM) classifier 25 / 25.0 pts
Q3: Softmax classifier 20 / 20.0 pts
Q4: Two-Layer Neural Network 25 / 25.0 pts
Q5: Image Features 10 / 10.0 pts
Q6: Bonus 0.0 pts
HW2 Score (out of 100): 94
HW2 Bonus (out of 10, if applicable): 2

Here's the breakdown of your score:
Fully-Connected Nets (25): 21
Batch Normalization (25): 25
Dropout (10): 8
ConvNets (30): 30
TensorFlow/PyTorch(10): 10
Bonus (Batch Norm Q2): 2
Bonus (Q6):
Comments: Q1: overfitting section incorrect (-4); Q3: Incorrect forward pass for inverted dropout (-2); 
HW3 Score (out of 100): 86
HW3 Bonus (out of 10, if applicable): 0

RNN_Captioning (25): 25
LSTM_Captioning (30): 29
Network Vis (15): 13
Style Transfer (15): 15
GAN (15): 4
Bonus LSTM:
Bonus GAN: 0
Comments: Q2: LSTM backward dWh (-1); Q3: Class Visualization: Partially incorrect application of gradients to the image.; Q5: Initial test (+2); GAN discriminator/generator # parameters (+2);
Normalized Project Grade (out of 100): 91.5

Comments from TA1: Nice writeup. You provide nice motivation for the problem. You tackle an interesting problem with a novel approach. However, in most examples when the region of interest does not encompass the majority of the image the bounding box is quite off the mark which is also revealed by your quantitative results. You are honest about the shortcoming and have a good discussion and analysis of results.

Comments from TA2: Good work - interesting project with a nice write-up and discussion.

Here are some stats:
Mean: 85.0
Median: 86.2
Std. Dev: 10.8

For the final project, refer to this

You can’t perform that action at this time.