Giter Club home page Giter Club logo

kittibox's Introduction

Hi there 👋

My name is Åsmund Brekke and I am currently a co-founder in Catchwise - a startup within the fishing industry.

I usually do backend, cloud, data engineering and AI stuff ☁️ I have also co-written two papers which you can find here and here 🎓

  • 💬 Buzzwords: .NET, Java, .py, .go, Azure, machine learning, k8s
  • 🌱 Currently looking into tools for the fishing industry ⛵
  • 📫 Reach me here

kittibox's People

Contributors

fredrik00 avatar marvinteichmann avatar nghiattran avatar villanuevab avatar

Watchers

avatar avatar avatar avatar

kittibox's Issues

Training for multiple classes

Thank you for your work! I tried using it for training for more classes ( 6 classes ) on Lyft Dataset. I updated input/bdd_input.py file to suit the needs of Lyft dataset. It also required changing hypes/kittiBox.json to reflect new image dimensions. (height: 1024, width: 1248)

While trying to train, I ran into a few weird issues.

  1. The loss function suddenly explodes and the program crashes with OverflowError: value too large to convert to int.

Here is the sample output from training (showing exploding loss):


`2020-06-30 03:21:22,589 root INFO Step 0/140000: loss = 39.73; lr = 1.00e-05; 0.104 sec (per Batch); 48.1 imgs/sec
2020-06-30 03:21:23,009 root INFO       (raw)  Acc.: 0.12, Conf: 31.51, Box: 2.22, Weight: 0.38, Delta: 2.61
2020-06-30 03:21:23,009 root INFO    (smooth)  Acc.: 0.12, Conf: 31.51, Box: 2.22, Weight: 0.38, Delta: 2.61
2020-06-30 03:21:23,915 root INFO Drawing encoded images (data utils)
 
2020-06-30 03:22:54,322 root INFO Step 200/140000: loss = 32.08; lr = 1.00e-05; 1.826 sec (per Batch); 2.7 imgs/sec
2020-06-30 03:22:54,481 root INFO       (raw)  Acc.: 0.76, Conf: 41.94, Box: 1.88, Weight: 0.38, Delta: 0.05
2020-06-30 03:22:54,481 root INFO    (smooth)  Acc.: 0.15, Conf: 32.03, Box: 2.20, Weight: 0.38, Delta: 2.48
2020-06-30 03:23:15,353 root INFO Drawing encoded images (data utils)
 
2020-06-30 03:24:26,125 root INFO Step 400/140000: loss = 772.73; lr = 1.00e-05; 1.268 sec (per Batch); 3.9 imgs/sec
2020-06-30 03:24:26,274 root INFO       (raw)  Acc.: 0.85, Conf: 28.31, Box: 1.34, Weight: 0.38, Delta: 1316.65
2020-06-30 03:24:26,275 root INFO    (smooth)  Acc.: 0.18, Conf: 31.85, Box: 2.16, Weight: 0.38, Delta: 68.19
2020-06-30 03:25:08,447 root INFO Drawing encoded images (data utils)
 
2020-06-30 03:25:59,540 root INFO Step 600/140000: loss = 399236.34; lr = 1.00e-05; 0.850 sec (per Batch); 5.9 imgs/sec
2020-06-30 03:25:59,712 root INFO       (raw)  Acc.: 0.85, Conf: 19.65, Box: 1.41, Weight: 0.38, Delta: 314026.66
2020-06-30 03:25:59,715 root INFO    (smooth)  Acc.: 0.22, Conf: 31.24, Box: 2.12, Weight: 0.38, Delta: 15766.12
2020-06-30 03:27:03,137 root INFO Drawing encoded images (data utils)
 
2020-06-30 03:27:32,453 root INFO Step 800/140000: loss = 19764840.00; lr = 1.00e-05; 0.419 sec (per Batch); 11.9 imgs/sec
2020-06-30 03:27:32,613 root INFO       (raw)  Acc.: 0.85, Conf: 23.55, Box: 1.84, Weight: 0.39, Delta: 21771650.00
2020-06-30 03:27:32,614 root INFO    (smooth)  Acc.: 0.25, Conf: 30.85, Box: 2.11, Weight: 0.38, Delta: 1103560.25
 
2020-06-30 03:28:57,191 root INFO Step 1000/140000: loss = 166476288.00; lr = 1.00e-05; 1.692 sec (per Batch); 3.0 imgs/sec
2020-06-30 03:28:57,340 root INFO       (raw)  Acc.: 0.85, Conf: 29.38, Box: 2.21, Weight: 0.39, Delta: 186277808.00
2020-06-30 03:28:57,341 root INFO    (smooth)  Acc.: 0.28, Conf: 30.78, Box: 2.11, Weight: 0.38, Delta: 10362273.00
2020-06-30 03:28:57,392 root INFO Drawing encoded images (data utils)`
  1. The bounding boxes start looking like dots in images denoting something is definitely wrong.

At 1st iteration, predicted boxes look normal:
predicted_boxes
1_pred

Ground_truth
1_true

At 3rd Iteration, predicted boxes start looking like dots.
predicted_boxes
5_pred

Ground_truth
5_true

Have you by any chance encountered something similar while dealing with the training.
Any insights into this from your expertise would be very useful.

Recommend Projects

  • React photo React

    A declarative, efficient, and flexible JavaScript library for building user interfaces.

  • Vue.js photo Vue.js

    🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.

  • Typescript photo Typescript

    TypeScript is a superset of JavaScript that compiles to clean JavaScript output.

  • TensorFlow photo TensorFlow

    An Open Source Machine Learning Framework for Everyone

  • Django photo Django

    The Web framework for perfectionists with deadlines.

  • D3 photo D3

    Bring data to life with SVG, Canvas and HTML. 📊📈🎉

Recommend Topics

  • javascript

    JavaScript (JS) is a lightweight interpreted programming language with first-class functions.

  • web

    Some thing interesting about web. New door for the world.

  • server

    A server is a program made to process requests and deliver data to clients.

  • Machine learning

    Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.

  • Game

    Some thing interesting about game, make everyone happy.

Recommend Org

  • Facebook photo Facebook

    We are working to build community through open source technology. NB: members must have two-factor auth.

  • Microsoft photo Microsoft

    Open source projects and samples from Microsoft.

  • Google photo Google

    Google ❤️ Open Source for everyone.

  • D3 photo D3

    Data-Driven Documents codes.