Skip to content

Latest commit

 

History

History
202 lines (187 loc) · 5.44 KB

README.md

File metadata and controls

202 lines (187 loc) · 5.44 KB
Raw

SPAIC

This repository is to collect progress of 60 days of Udacity challenge.

I was having my board exams lately so it was harder for me to collect everything i studied here.
Day Task Links Remarks
1-20 Followed Udacity's Secure and Private AI Challange Need to revise.
21 DCR Numerals Using FFN Accuracy very low. Model needs more work.
22 DCR Numerals Using CNN Miracle happened. Accuracy 90+
23 Tried to understand yesterday's miracle. Wrote some CNN codes to optimize. Just another failure try.
24 Also i watched Federated learning with trusted aggregator video. Today i used the log_softmax function on final layer of my CNN, then the problem of accuracy solved. Yesterday's miracle was due to randomization of parameters.
25 Trained DCR all characters using Pytorch. Plus I learned about deep dream. Used the final function log_softmax and everything solved.
26 Did lots of task. Downloaded, processed and trained a CNN for smog detection. Datasets Notebook It was tough day and all training done on local computer. Took whole day. Model have accuracy 89% upto 10 epochs.
27 Trained Smog Detection datasets on Colab Notebook Used custom VGG
28 - 32 Finished Course - Need to rewatch
33 Basics of Flask and RESTful API - To deploy ML model into web app
34 Virtual Meetups and little more Flask - Flask is tough for me.
35 #flask, #django and #MachineLearning #engineer pathway on #AWS student, enrolled in #AWSDeepRacer, learned AWS DeepLens, collected datset, papers for people counting problem - AWS is amazing but without mastercard little i could do
36 Worked on project 'Crowd Density Detection', Wrote lines of codes to just see memory error but at the end of day trained a simple #regression model (overfitted) - Crowd density detection is harder than smog Detection
37 * Did little search about crowd density estimation * Improved little accuracy of Smog Detection https://t.co/mwc9aP1K7M?amp=1
38 Studied about Artistic Neural Style Transfer and coded it * https://t.co/duiEspN3xk?amp=1 * https://t.co/auwso3IeT6?amp=1 If i am ever going to make a movie, then i will use Neural Networks for scene creation.
39 * Math behind Gaussian Kernel Mapping(for Crowd Density Estimation) * Learned about various object tracking(including YOLO) history from Siraj Raval's Video(best) Siraj's Videos are coolest one.
40 * Learned even more about YOLO * Worked on custom Smog/Clear Dataset for Smog Detection - Dataset was named Smog4000 by group members.
41 * Studied object detection with YOLO and opencv * Learned about centroid tracking algorithm and several more in #opencv - Object tracking is awesome.
42-43 * A long streak of debugging the code * Learning real world object tracking(w speed) using YOLO and Centroid tracking algorithm - -
44 * Improved a segmentation code for devanagari character recognition using #numpy * Learning real world object tracking(w speed) using YOLO and Centroid tracking algorithm Numpy is gold.
45-46 * Busy days due to many study projects on going * Learning real world object tracking(w speed) using YOLO and Centroid tracking algorithm * Writing google doc for Crowd Density Estimation -
47 * Crowd density article completed. * Starting RASA -

Projects I was/am involved during 60 days of Udacity.

  • Devanagari Handwritten Character/Word Recognition: Please follow the link to check out the main Devanagari Handwritten Character and Word recognition.
  • Smog Detection. Here