Scania_Thesis

Code for Semantic Segmentation

Analysis

Contribution to the understanding of the inference and evaluate the results
Inference and visualization for the Digits
Compute PCA and confusion matrix

Annotation

Generate labels from XML files using hard-coded color codes

Augmentation

Generate randomly cropped and rotated images

DNN Inference

Caffe model inference without the use of Digits

Plots and Tests

Old tests used for analysis

Steps To Remember

Drivenet Example

cd /usr/local/driveworks-0.2.1/bin/
./sample_drivenet --input-type=video --video=/media/nvidia/iQMatic_2/Scania_truck_recording/2017-03-10_DrivePX2_HDL32E_Gläntan_Etc/20170310_Camera/scania_0005.h264
./sample_camera_gmsl --input-type=camera --camera-type=c-ov10640-b1 --csi-port=ab --camera-index=0
./sample_lidar_replay --device=VELO_HDL32E --ip=192.168.1.60 --port=2368 --scan-frequency=10

cd /usr/local/driveworks-0.2.1/tools
sudo ./recorder --config-file=../data/tools/glantan.json

./replayer --folder=../data/recordings/dw_20170316_110745_0.000000_0.000000

Recording for Bus

Update JSON file

sudo ./recorder --config-file=../data/tools/glantan_ssd.json

Set IP for lidar

ip l
sudo ifconfig eth0 192.168.1.60
sudo ifconfig eth0 down
sudo ifconfig eth0 up

Feature Extraction

./sample_camera_tracker --video=/home/scania/Scania/Glantan_Recordings/2017-03-24_DrivePX2/dw_20170324_115428_0.000000_0.000000/video_front.h264
cd /Scania/Opencv/samples/python
python video_gabor_read.py

CAFFE NET

cd /home/scania/Scania/Deep_Learning/caffe-segnet-cudnn5/build/tools/
./caffe train -gpu 0 -solver /home/scania/Scania/Deep_Learning/SegNet-Tutorial/Models/segnet_solver.prototxt  # This will begin training SegNet on GPU 0
./caffe train -gpu 0 -solver /home/scania/Scania/Deep_Learning/SegNet-Tutorial/Models/segnet_basic_solver.prototxt  # This will begin training SegNet-Basic on GPU 1
./caffe train -gpu 0 -solver /home/scania/Scania/Deep_Learning/SegNet-Tutorial/Models/segnet_solver.prototxt -weights /home/scania/Scania/Deep_Learning/SegNet-Tutorial/Models/VGG_ILSVRC_16_layers.caffemodel  # This will begin training SegNet on GPU 0 with a pretrained encoder

python /home/scania/Scania/Deep_Learning/SegNet-Tutorial/Scripts/compute_bn_statistics.py /home/scania/Scania/Deep_Learning/SegNet-Tutorial/Models/segnet_train.prototxt /home/scania/Scania/Deep_Learning/SegNet-Tutorial/Models/Training/segnet_iter_1000.caffemodel /home/scania/Scania/Deep_Learning/SegNet-Tutorial/Models/Inference/ 

python /home/scania/Scania/Deep_Learning/SegNet-Tutorial/Scripts/test_segmentation_camvid.py --model /home/scania/Scania/Deep_Learning/SegNet-Tutorial/Models/segnet_inference.prototxt --weights /home/scania/Scania/Deep_Learning/SegNet-Tutorial/Models/Inference/test_weights_segnet_2.caffemodel --iter 10

python /home/scania/Scania/Deep_Learning/SegNet-Tutorial/Scripts/test_segmentation_camvid.py --model /home/scania/Scania/Deep_Learning/SegNet-Tutorial/Models/segnet_basic_inference.prototxt --weights /home/scania/Scania/Deep_Learning/SegNet-Tutorial/Models/Inference/test_weights_segnet_basic_2.caffemodel --iter 10  # Test SegNetBasic

https://gist.github.com/weiliu89/2ed6e13bfd5b57cf81d6
https://gist.github.com/weiliu89/45e9e8de2c13af6476ca#file-readme-md
train --solver=/home/scania/Scania/Deep_Learning/digits/digits/jobs/20170406-093958-021f/solver.prototxt --gpu=0 --weights=/home/scania/Scania/Deep_Learning/digits/examples/semantic-segmentation/VGG_ILSVRC_16_layers_fc_reduced.caffemodel"

Get Images From Video

ffmpeg -i scania_0001.raw qscale:v 2 output_%04.jpg
ffmpeg -i video_front.h264 -r 25 20170324_115428_%04d.bmp 																# Best to use
avconv -i video_front.h264 20170324_113202_%04d.jpg
avconv -i video_front.raw -vcodec copy -acodec copy output_front.h264  //FAILED
ffmpeg -r 5 -start_number 1001 -i 20170324_115428_%04d.png -c:v libx264 -vf fps=25 -pix_fmt yuv420p out.mp4				# Images to video

Resize Image

sudo apt-get install imagemagick
convert  -resize 50% source.png dest.jpg
convert  -resize 1024X768  source.png dest.jpg

mogrify -resize 50% *png      											# keep image aspect ratio
mogrify -resize 480x360 *jpg  											# keep image aspect ratio
mogrify -resize 480x360! *png 											# don't keep image aspect ratio
mogrify -resize x240 *png     											# don't keep image aspect ratio
mogrify -resize 320x *png     											# don't keep image aspect ratio
mogrify -format png *.jpg	  											# convert jpg to png type
mogrify -format jpg *.png	  											# convert png to jpg type
mogrify -flop -write *_R.png  											# flop the image
convert *.jpg -flop -set filename:new "%t_R" "%[filename:new].jpg"   	# Rotate image and save with _R

Presentation 4 April

cd /usr/local/driveworks-0.2.1/tools

# Scania gate to Glantan gate vertical stand
./replayer --folder=/home/scania/Scania/Glantan_Recordings/2017-03-24_DrivePX2/dw_20170324_114545_0.000000_0.000000

# Gravel pile
./replayer --folder=/home/scania/Scania/Glantan_Recordings/2017-03-24_DrivePX2/dw_20170324_120648_0.000000_0.000000

# Return to Scania horizontal
./replayer --folder=/home/scania/Scania/Glantan_Recordings/2017-03-24_DrivePX2/dw_20170324_114545_0.000000_0.000000

# Replay video 11
cd /usr/local/driveworks-0.2.1/bin
./sample_camera_replay --video=/home/scania/Scania/Glantan_Recordings/2017-03-10_Camera/output_0011.h264
./sample_lidar_replay --file=/home/scania/Scania/Glantan_Recordings/2017-03-10_Camera/scania_0011.bin

Inference

cd Scania/Deep_Learning/digits/digits/tools/
python inference_agneev.py /home/scania/Scania/Deep_Learning/DataSets/Test_Data/scania_sample/image-test-inference.txt /home/scania/Scania/Deep_Learning/DataSets/Test_Data/scania_sample/Inference 20170406-131719-c009
python inference_agneev.py /home/scania/Scania/Glantan_Recordings/2017-03-24_DrivePX2/dw_20170324_115921_0.000000_0.000000/video_front.h264 /home/scania/Scania/Deep_Learning/DataSets/Test_Data/scania_sample/Inference 20170406-131719-c009

TO DO

Confusion matrix https://github.com/pandas-ml/pandas-ml http://pandas-ml.readthedocs.io/en/stable/conf_mat.html http://scikit-learn.org/stable/auto_examples/model_selection/plot_confusion_matrix.html#sphx-glr-auto-examples-model-selection-plot-confusion-matrix-py

Overnight Test

Remove +30 options
python inference_agneev.py /home/scania/Scania/Glantan_Recordings/2017-04-07_DrivePX2/20170405_124731/video_front.h264 /home/scania/Scania/Deep_Learning/DataSets/Test_Data/scania_sample/Inference 20170419-213931-6881
python inference_agneev.py /home/scania/Scania/Agneev/Labels/Images_List.txt /home/scania/Scania/Deep_Learning/DataSets/Test_Data/scania_sample/Inference 20170425-115817-66fd
/home/scania/Scania/Glantan_Recordings/2017-04-07_DrivePX2/dw_20170405_123343_0.000000_0.000000/video_front.h264

Job Numbers

CLASSES	DATASET	MODELS
7 classes:	20170419-152029-5e0d	20170419-153352-2f2f
16 classes:	20170419-213812-86a7	20170419-213931-6881
9 classes:	20170426-114517-ceba	20170426-114600-520a
10 classes:	20170426-144243-3e52	20170426-144433-798d
5 classes:	20170426-150630-756f	20170426-150713-36b9
===	===	===
16 classes_682	20170503-010848-a41c	20170503-011008-dc7d
9 classes_682	20170503-044312-aa42	20170503-045532-da28
===	===	===
16 classes_1364	===	===
9 classes_1364	20170515-155318-1a09	===

Digits Inference

cd Scania/Deep_Learning/digits/digits/tools/
python inference_agneev.py /home/scania/Scania/Agneev/Labels/Images_List.txt /home/scania/Scania/Deep_Learning/DataSets/Test_Data/scania_sample/Inference 20170503-045532-da28

Build DriveNet example

cd /usr/local/driveworks-0.2.1/samples/build
sudo cmake -DCMAKE_BUILD_TYPE=Release /usr/local/driveworks-0.2.1/samples/src/dnn/sample_object_detector

Object Detector Test

cd /usr/local/driveworks-0.2.1/samples/src/dnn/sample_object_detector
./sample_object_detector --caffe_model=/home/scania/Scania/Agneev/DNN_Data/Agneev_16/snapshot_iter_16380.caffemodel --caffe_prototxt=/home/scania/Scania/Agneev/DNN_Data/Agneev_16/deploy.prototxt --video=/home/scania/Scania/Glantan_Recordings/2017-04-07_DrivePX2/dw_20170405_124107_0.000000_0.000000/video_front.h264

Python Segnet Test

cd ~/Scania/Agneev/Code/DNN_Inference
python test_segnet.py --model=/home/scania/Scania/Agneev/DNN_Data/Agneev_16/deploy.prototxt --weights=/home/scania/Scania/Agneev/DNN_Data/Agneev_16/snapshot_iter_16380.caffemodel --colours=/home/scania/Scania/Agneev/Code/DNN_Inference/lut_16.png --data=/home/scania/Scania/Agneev/Code/DNN_Inference/Images/Images.txt

Segnet Presentation for Demo

~/Scania/Agneev/Results/Half_Time_Presentation_Videos
python test_segnet.py --model=Agneev_16/deploy.prototxt --weights=Agneev_16/snapshot_iter_16380.caffemodel --colours=Agneev_16/lut_16.png --data=Video_Set_4/Images_List.txt
python test_segnet.py --model=Agneev_16/deploy.prototxt --weights=Agneev_16/snapshot_iter_16380.caffemodel --colours=Agneev_16/lut_16.png --data=Video_Set_4/Images_List_with_mask.txt

To obtain New Color Labels

Place files in /home/scania/Scania/Agneev/Labels/Color_Labels/

cd /home/scania/Scania/Agneev/Code/Augmentation
python flip_image.py

cd /home/scania/Scania/Agneev/Code/Calibration
python save_undistorted.py

cd /home/scania/Scania/Agneev/Code/Augmentation
random_cropping_rotation.py

To split data to training, validation and test sets

cd /home/scania/Scania/Agneev/Code/Split_Data
python shuffle_split_files.py

Name		Name	Last commit message	Last commit date
Latest commit History 22 Commits
Analysis		Analysis
Annotation		Annotation
Augmentation		Augmentation
Calibration		Calibration
DNN_Inference		DNN_Inference
Matlab		Matlab
Optical_Flow		Optical_Flow
Plots_and_Tests		Plots_and_Tests
Result_XLS_Files		Result_XLS_Files
Split_Data		Split_Data
Installations		Installations
README.md		README.md

agneevguin/Scania_Thesis

Folders and files

Latest commit

History

Repository files navigation

Scania_Thesis

Code for Semantic Segmentation

Analysis

Annotation

Augmentation

DNN Inference

Plots and Tests

Steps To Remember

Drivenet Example

Recording for Bus

Set IP for lidar

Feature Extraction

CAFFE NET

Get Images From Video

Resize Image

Presentation 4 April

Inference

TO DO

Overnight Test

Job Numbers

Digits Inference

Build DriveNet example

Object Detector Test

Python Segnet Test

Segnet Presentation for Demo

To obtain New Color Labels

To split data to training, validation and test sets

About

Resources

Stars

Watchers

Forks

Languages