Journal.Datasets
Used:
Not Used:
check this out https://scale.com/open-datasets
After a short investigation, we won't use this method yet
ICRA 2020
KITTI-2011
├── 2011_09_30_drive_0018_sync
├── 2011_09_30_drive_0020_sync
├── 2011_09_30_drive_0027_sync
├── 2011_09_30_drive_0028_sync
├── 2011_09_30_drive_0033_sync
├── 2011_09_30_drive_0034_sync
├── 2011_10_03_drive_0027_sync
└── 2011_10_03_drive_0034_sync
search for other seq? redo the labelling?
KITTI-360
├── 2013_05_28_drive_0000_sync
├── 2013_05_28_drive_0002_sync
├── 2013_05_28_drive_0003_sync
├── 2013_05_28_drive_0004_sync
├── 2013_05_28_drive_0005_sync
├── 2013_05_28_drive_0006_sync
├── 2013_05_28_drive_0007_sync
├── 2013_05_28_drive_0009_sync
└── 2013_05_28_drive_0010_sync
FROM i7g4 /media/RAIDONE/DATASETS/Apolloscapes/Road11/image/GZ20180310B
Documentation Beijing, China
Sequences have huge roads with avenues, too big
[STEREO + LIDAR]
65 sequences in the training for 3d tracking task , divided in 4 compressed parts
FROM i7g4
/media/RAIDONE/DATASETS/ARGO/only_center_camera/train1
/media/RAIDONE/DATASETS/ARGO/only_center_camera/train2
/media/RAIDONE/DATASETS/ARGO/only_center_camera/train3
/media/RAIDONE/DATASETS/ARGO/only_center_camera/train4
Images are from Miami and Pittsburgh. Some intersections are huge but someones in Miami seem ok.
from i7g4 /media/RAIDONE/DATASETS/oxford_img_bb
"all-year" long dataset, but the number of intersections is reduced
Downloaded file is train.tar from perception dataset
/media/ballardini/7D3AD71E1EACC626/DATASET/JournalVersion/LYFT/
** PNG to MP4
Use generate.videos.bash
Even though some intersections are present, most of the sequences are in big avenues; not worth it
San Francisco; 158 short sequences, 4 seconds each
Downloaded
pandaset_pandaset_0.zip
pandaset_pandaset_1.zip
pandaset_pandaset_2.zip
from Boston and Singapore - nuscenes 10 parts, US or ASIA should be just mirrors, even thought some compressed folders have different names.
v1.0-trainval01_blobs_camera.tgz /media/ballardini/7D3AD71E1EACC626/DATASET/JournalVersion/NUSCENES/
Might be interesting for the TYPE-6 , 4-way, crossings. Long sequences + Boston seems good.
(website)[https://irap.kaist.ac.kr/dataset/]
not too many small/acceptable intersections, (see maps here)[https://irap.kaist.ac.kr/dataset/download_2.html]
Images from raw files are 1920x1200 - resize as needed.
** RAW to MP4
ffmpeg -f rawvideo -pixel_format bayer_rggb8 -video_size 1920x1200 -framerate 10 -i filename.raw -vf scale=640:-1 -c:v libx264 filename.mp4
** Extract PNG from RAW
ffmpeg -f rawvideo -pixel_format bayer_rggb8 -video_size 1920x1200 -framerate 10 -i filename.raw -vf scale=640:-1 folder/%010d.png