Commits

Branch selector

 master

User selector

All users

Datepicker

All time

Commit History

Commits on Aug 30, 2016

• Remove debug print statement accidentally left from earlier commit

  matthewearl

  committed
  db823eb

  Copy full SHA for db823eb

Commits on Aug 29, 2016

• Merge branch 'jackmiller1-multiple-fonts'. Fixes matthewearl#1.

  matthewearl

  committed
  8e6acb1

  Copy full SHA for 8e6acb1

• Update README.md to with fonts/ folder information

  matthewearl

  committed
  03fe13a

  Copy full SHA for 03fe13a

• Stop font chars being regenerated each batch by making gen.generate_ims() generate an infinite number of images, and making train.gen_batches() use a single instance of gen.generate_ims()

  matthewearl

  committed
  5ea9421

  Copy full SHA for 5ea9421

• Merge branch 'multiple-fonts' of https://github.com/jackmiller1/deep-anpr into jackmiller1-multiple-fonts

  matthewearl

  committed
  4a71ef9

  Copy full SHA for 4a71ef9

Commits on Aug 25, 2016

• Merge branch 'master' of https://github.com/matthewearl/deep-anpr

  matthewearl

  committed
  fe4d736

  Copy full SHA for fe4d736

• Fixes matthewearl#7: Don't backprop on character probabilities where the number plate isn't present. Do this by setting the digit_loss on characters corresponding with non-present plates to 0.

  matthewearl

  committed
  cbeb02d

  Copy full SHA for cbeb02d

Commits on Jun 16, 2016

• randomly select a font from './fonts' folder for each image

  

  Jack Miller

  committed
  44da6f1

  Copy full SHA for 44da6f1

Commits on May 26, 2016

• Update README with paragraph about practicality, and how long training takes

  matthewearl

  committed
  4b781d5

  Copy full SHA for 4b781d5

Commits on May 7, 2016

• Add a note about extractbgs taking a while to run

  matthewearl

  committed
  10fd4c0

  Copy full SHA for 10fd4c0

• Fix SUN DB file size in README (36GB not 39GB)

  matthewearl

  committed
  ac1ef87

  Copy full SHA for ac1ef87

• Fix SUN397.tar.gz link in README, and add shebangs to executables

  matthewearl

  committed
  a419ffd

  Copy full SHA for a419ffd

• Make an extractbgs script, have extractbgs and gen create their target directories, and make the instructions clearer

  matthewearl

  committed
  fdad36b

  Copy full SHA for fdad36b

Commits on May 6, 2016

• Fix link to blog post in README.md

  matthewearl

  committed
  7f0be94

  Copy full SHA for 7f0be94

• Add a README.md

  matthewearl

  committed
  71e61ac

  Copy full SHA for 71e61ac

Commits on May 1, 2016

• Add docstrings and general tidy-up

  matthewearl

  committed
  3b47094

  Copy full SHA for 3b47094

• Add copyright notice to files without one

  matthewearl

  committed
  6a00c52

  Copy full SHA for 6a00c52

• Use non-maximum suppression in detect.py based on the presence output from the network

  matthewearl

  committed
  48af712

  Copy full SHA for 48af712

• Vary yaw by a bit more

  matthewearl

  committed
  f9a7b3c

  Copy full SHA for f9a7b3c

Commits on Apr 27, 2016

• Add post-processing step to detect.py, to coalesence overlapping rectangles

  matthewearl

  committed
  2842842

  Copy full SHA for 2842842

Commits on Apr 26, 2016

• detect.py: Use a higher presence threshold to eliminate false positives

  matthewearl

  committed
  a1f3920

  Copy full SHA for a1f3920

Commits on Apr 17, 2016

• detect.py: Filter based on digit certainty, as well as the presence indicator

  matthewearl

  committed
  271f2ec

  Copy full SHA for 271f2ec

Commits on Apr 15, 2016

• Fix training: Images input to the NN should be in 2d, rather than flattened

  matthewearl

  committed
  c5c066f

  Copy full SHA for c5c066f

• Fix bugs in detector. TBD: Fix regressions in trainer

  matthewearl

  committed
  a673be7

  Copy full SHA for a673be7

Commits on Apr 14, 2016

• Initial commit of detect.py

  matthewearl

  committed
  5d8e4f1

  Copy full SHA for 5d8e4f1

Commits on Apr 13, 2016

• Remove more dead detector code

  matthewearl

  committed
  ce50cea

  Copy full SHA for ce50cea

• Simplify output layer of training model

  matthewearl

  committed
  7cc0dfa

  Copy full SHA for 7cc0dfa

• Remove unused 'detector' model, which was intended to be used as the first phase in a cascade to increase detection speed. This may be revived in the future.

  matthewearl

  committed
  10debba

  Copy full SHA for 10debba

• Move model into its own module

  matthewearl

  committed
  0de6632

  Copy full SHA for 0de6632

Commits on Apr 12, 2016

• Remove redundant threadgen decorator

  matthewearl

  committed
  c5545f2

  Copy full SHA for c5545f2

• Various optimizations

  matthewearl

  committed
  beb184a

  Copy full SHA for beb184a

• Fix gen.py paths

  matthewearl

  committed
  3294567

  Copy full SHA for 3294567

Commits on Apr 7, 2016

• Increase size of model to more closely match that of Stark et al, fix bug in loss function related to broadcasting, and improve reports

  matthewearl

  committed
  7702f2e

  Copy full SHA for 7702f2e

Commits on Apr 4, 2016

• Add of a script for visualizing the first layer of weights

  matthewearl

  committed
  8cb4792

  Copy full SHA for 8cb4792

• Experiments with making the system detect presence of a plate, as well as decoding it

  matthewearl

  committed
  88ea28f

  Copy full SHA for 88ea28f

Pagination

Previous

Next