FRRN Tensorflow

Intro

This repository contains code implementing the paper, Full Resolution Residual Networks for Semantic Segmentation in Street Scenes (FRRN) in Tensorflow.

:warning: This is not an official implementation, and might have some glitch (,or a major defect).

Requirements

Python 3.5.2 (or higher)
Tensorflow v1.3 (or higher)
Numpy
PIL,Matplotlib,numpy,tqdm,better_exceptions,...

Training from scratch

Dataset preprocessing

Donwload datasets(gtFine_trainvaltest.zip, leftImg8bit_trainvaltest.zip) and unzip under the directory dataset
- Make sure that path look like ./datasets/cityspaces/gtFine and ./datasets/cityspaces/leftImg8bit.
Run the script to generate TrainId format images.
- python cityscapesScripts/cityscapesscripts/preparation/createTrainIdLabelImgs.py

Run training

Run, python main.py
If you want to change some hyperparameters, then check the config dictionary on the main.py.

Run evaluation

Comment out the main() at the second last line on the main.py and uncomment the last line, eval().
Then, Run python main.py.

Trained Results

The training is done with nVidia Quadro M4000 GPU (8GB of V-RAM). I can fit 3 batches with Type A architecture. I ran this about 32 hours for about 55k iterations. Below is the learning statistics, and some results on training set.

Training statistics

training_scalars

training_images

Result on the Validation Set

Below is the validation set result. I could get mIOU of 0.570.

❯❯❯ python cityscapesScripts/cityscapesscripts/evaluation/evalPixelLevelSemanticLabeling.py

Evaluating 500 pairs of images...
Images Processed: 500

-------------- ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ -------
              |  u   |  e   |  r   |  o   |  s   |  d   |  g   |  r   |  s   |  p   |  r   |  b   |  w   |  f   |  g   |  b   |  t   |  p   |  p   |  t   |  t   |  v   |  t   |  s   |  p   |  r   |  c   |  t   |  b   |  c   |  t   |  t   |  m   |  b   | Prior |
-------------- ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ -------
    unlabeled | 0.01   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.10   0.00   0.00   0.00   0.00   0.00   0.01   0.00   0.00   0.00   0.03   0.00   0.84   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00  0.0004
  ego vehicle | 0.90   0.00   0.00   0.00   0.00   0.00   0.00   0.10   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00  0.0490
rectification | 0.86   0.00   0.00   0.00   0.00   0.00   0.00   0.02   0.00   0.00   0.00   0.07   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.01   0.02   0.00   0.01   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00  0.0189
   out of roi | 0.39   0.00   0.00   0.00   0.00   0.00   0.00   0.14   0.04   0.00   0.00   0.18   0.00   0.00   0.00   0.00   0.00   0.01   0.00   0.00   0.01   0.13   0.00   0.05   0.00   0.00   0.04   0.00   0.00   0.00   0.00   0.00   0.00   0.00  0.0151
       static | 0.28   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.03   0.00   0.00   0.39   0.00   0.02   0.00   0.00   0.00   0.05   0.00   0.00   0.07   0.08   0.00   0.01   0.01   0.00   0.03   0.00   0.00   0.00   0.00   0.00   0.00   0.02  0.0149
      dynamic | 0.33   0.00   0.00   0.00   0.00   0.00   0.00   0.01   0.02   0.00   0.00   0.35   0.00   0.03   0.00   0.00   0.00   0.02   0.00   0.00   0.01   0.03   0.00   0.01   0.06   0.01   0.06   0.00   0.00   0.00   0.00   0.00   0.02   0.04  0.0042
       ground | 0.03   0.00   0.00   0.00   0.00   0.00   0.00   0.54   0.39   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.03   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00  0.0178
         road | 0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.98   0.01   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00  0.3293
     sidewalk | 0.02   0.00   0.00   0.00   0.00   0.00   0.00   0.09   0.85   0.00   0.00   0.01   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.01   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00  0.0473
      parking | 0.25   0.00   0.00   0.00   0.00   0.00   0.00   0.51   0.18   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.01   0.00   0.00   0.00   0.04   0.00   0.00   0.00   0.00   0.00   0.00   0.00  0.0040
   rail track | 0.18   0.00   0.00   0.00   0.00   0.00   0.00   0.67   0.11   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.03   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00  0.0006
     building | 0.01   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.93   0.00   0.00   0.00   0.00   0.00   0.01   0.00   0.00   0.01   0.02   0.00   0.01   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00  0.1917
         wall | 0.12   0.00   0.00   0.00   0.00   0.00   0.00   0.06   0.09   0.00   0.00   0.28   0.22   0.10   0.00   0.00   0.00   0.01   0.00   0.00   0.00   0.09   0.01   0.00   0.01   0.00   0.01   0.00   0.00   0.00   0.00   0.00   0.00   0.00  0.0064
        fence | 0.10   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.02   0.00   0.00   0.27   0.01   0.47   0.00   0.00   0.00   0.03   0.00   0.00   0.01   0.06   0.00   0.00   0.01   0.00   0.01   0.00   0.00   0.00   0.00   0.00   0.00   0.01  0.0072
   guard rail | 0.34   0.00   0.00   0.00   0.00   0.00   0.00   0.01   0.10   0.00   0.00   0.00   0.00   0.27   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.03   0.21   0.00   0.00   0.00   0.04   0.00   0.00   0.00   0.00   0.00   0.00   0.00  0.0000
       bridge | 0.52   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.32   0.00   0.04   0.00   0.00   0.00   0.01   0.00   0.00   0.00   0.09   0.00   0.00   0.00   0.00   0.01   0.00   0.00   0.00   0.00   0.00   0.00   0.00  0.0003
         pole | 0.04   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.01   0.00   0.00   0.17   0.00   0.01   0.00   0.00   0.00   0.63   0.00   0.00   0.01   0.09   0.00   0.00   0.01   0.00   0.01   0.00   0.00   0.00   0.00   0.00   0.00   0.01  0.0129
    polegroup | 0.14   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.04   0.00   0.00   0.21   0.00   0.22   0.00   0.00   0.00   0.23   0.00   0.00   0.01   0.03   0.01   0.00   0.05   0.00   0.01   0.00   0.00   0.00   0.00   0.00   0.00   0.04  0.0001
traffic light | 0.07   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.29   0.00   0.00   0.00   0.00   0.00   0.07   0.00   0.38   0.02   0.15   0.00   0.01   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00  0.0017
 traffic sign | 0.04   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.08   0.00   0.01   0.00   0.00   0.00   0.01   0.00   0.00   0.80   0.04   0.00   0.00   0.00   0.00   0.01   0.00   0.00   0.00   0.00   0.00   0.00   0.00  0.0058
   vegetation | 0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.02   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.96   0.01   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00  0.1515
      terrain | 0.03   0.00   0.00   0.00   0.00   0.00   0.00   0.04   0.14   0.00   0.00   0.01   0.01   0.01   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.12   0.64   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00  0.0073
          sky | 0.01   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.02   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.02   0.00   0.94   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00  0.0293
       person | 0.02   0.00   0.00   0.00   0.00   0.00   0.00   0.01   0.01   0.00   0.00   0.05   0.00   0.00   0.00   0.00   0.00   0.01   0.00   0.00   0.00   0.01   0.00   0.00   0.83   0.02   0.02   0.00   0.00   0.00   0.00   0.00   0.00   0.01  0.0114
        rider | 0.01   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.03   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.02   0.00   0.00   0.17   0.59   0.04   0.00   0.00   0.00   0.00   0.00   0.04   0.11  0.0019
          car | 0.01   0.00   0.00   0.00   0.00   0.00   0.00   0.01   0.00   0.00   0.00   0.01   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.01   0.00   0.00   0.00   0.00   0.96   0.00   0.00   0.00   0.00   0.00   0.00   0.00  0.0570
        truck | 0.06   0.00   0.00   0.00   0.00   0.00   0.00   0.01   0.00   0.00   0.00   0.07   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.01   0.01   0.00   0.05   0.00   0.00   0.29   0.49   0.00   0.00   0.00   0.00   0.00   0.00  0.0026
          bus | 0.07   0.00   0.00   0.00   0.00   0.00   0.00   0.01   0.00   0.00   0.00   0.07   0.00   0.01   0.00   0.00   0.00   0.00   0.00   0.00   0.03   0.02   0.00   0.00   0.00   0.00   0.24   0.21   0.31   0.00   0.00   0.03   0.00   0.00  0.0034
      caravan | 0.12   0.00   0.00   0.00   0.00   0.00   0.00   0.01   0.00   0.00   0.00   0.09   0.00   0.00   0.00   0.00   0.00   0.01   0.00   0.00   0.00   0.05   0.00   0.00   0.00   0.00   0.19   0.53   0.00   0.00   0.00   0.00   0.00   0.00  0.0001
      trailer | 0.32   0.00   0.00   0.00   0.00   0.00   0.00   0.01   0.01   0.00   0.00   0.12   0.00   0.00   0.00   0.00   0.00   0.01   0.00   0.00   0.00   0.03   0.00   0.00   0.00   0.00   0.45   0.05   0.00   0.00   0.00   0.00   0.01   0.00  0.0002
        train | 0.06   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.29   0.00   0.01   0.00   0.00   0.00   0.01   0.00   0.00   0.00   0.03   0.00   0.01   0.00   0.00   0.03   0.02   0.23   0.00   0.00   0.31   0.00   0.00  0.0010
   motorcycle | 0.05   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.01   0.00   0.00   0.03   0.00   0.01   0.00   0.00   0.00   0.01   0.00   0.00   0.00   0.01   0.00   0.00   0.04   0.05   0.26   0.00   0.00   0.00   0.00   0.00   0.36   0.17  0.0007
      bicycle | 0.02   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.01   0.00   0.00   0.03   0.00   0.01   0.00   0.00   0.00   0.01   0.00   0.00   0.00   0.01   0.00   0.00   0.02   0.02   0.04   0.00   0.00   0.00   0.00   0.00   0.03   0.79  0.0062
-------------- ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ -------

classes          IoU      nIoU
--------------------------------
road          : 0.963      nan
sidewalk      : 0.736      nan
building      : 0.873      nan
wall          : 0.209      nan
fence         : 0.382      nan
pole          : 0.515      nan
traffic light : 0.359      nan
traffic sign  : 0.617      nan
vegetation    : 0.898      nan
terrain       : 0.527      nan
sky           : 0.884      nan
person        : 0.714    0.522
rider         : 0.449    0.339
car           : 0.896    0.805
truck         : 0.349    0.194
bus           : 0.289    0.197
train         : 0.277    0.132
motorcycle    : 0.239    0.155
bicycle       : 0.658    0.507
--------------------------------
Score Average : 0.570    0.356
--------------------------------

categories       IoU      nIoU
--------------------------------
flat          : 0.975      nan
nature        : 0.900      nan
object        : 0.560      nan
sky           : 0.884      nan
construction  : 0.875      nan
human         : 0.739    0.563
vehicle       : 0.891    0.788
--------------------------------
Score Average : 0.832    0.675
--------------------------------

Notes (Implementation Differences)

Only implmented the Type A architecture; but it is very straight forward to implement the other version.
I randomly cropped the image of 1/4 area from original images and then resize images by its width and height by half for training.
- Therefore, the prediction is done by half scale, unlike the paper where prediction is made at 1/4 scale with Type A architecture.
- No translation-based augmentation is made. (Beta-augmentation is implmented.)
- By doing that, I keep the same input size to the paper, while prediction scale is doubled to the paper. It can be done since it is fully convolutional architecture.
- Nevertheless, it might be the reason why the trained model is worse than the author could get.
- Global context, maybe is more important for fine-grained predictions.
No separated gradient descent step is required.
- I can barely fit up to 3 batches on 8GB V-Ram with Type A.
About hyperparams:
- No learning rate deacy is applied.
- K could be a value to be tuned carefully.
Pretrained model is available upon request.

Acknowledgement

Author's original implementation in Theano with Lasagnelink

hiwonjoon / tf-frrn

readme