- About the network architecture of depth prediction, pose estimation and semantic segmentation, I make an analysis. In the future, I plan to build a SLAM system based all the information.
- Paper reading: Towards Better Generalization: Joint Depth-Pose Learning without PoseNet. About my understand of the details, I write a note. All the … Continue Reading ››
zhou のすべての投稿
先週の進捗(ZHOU)
- Read some paper and make a schedule about future research work. all the details are upload to gitlab.
- Paper submitted to the journal AROB is accepted.
先週の進捗(ZHOU)
Read some paper introducing the application of SLAM. About some application which I interested, I need to train the network on datasets which created by myself. There are many basic work need to been done before researching SLAM. Therefore, I decided to continue the research of SLAM in automated driving and hope it … Continue Reading ››
先週の進捗(ZHOU)
- Finished the modification of journal AROB, and update to Gitlab.
- Learn to arrange image in Latex like other paper, so that the original paper can be saved directly from the .pdf file.
先週の進捗(ZHOU)
- modify the paper of AROB journal, and still need some time to better show the result.
- paper read: Depth Estimation with Occlusion Modeling Using Light-Field Cameras
- according to the orientation of occlusion boundary, modified angular photo-consistency
- separating into two regions: occlusion and non-occlusion
- high accuracy, but the image resolution from the light-field camera is low
- Unfortunately, I … Continue Reading ››
先週の進捗(ZHOU)
- Finished the presentation in ICIAE2021 and 2020-IC-YZU-KYUTECH-CSLG;
- Configured the Latex For JIIAE and modify the paper, I will submit it today;
- Received some reviews of AROB journal and I will modify it and submit it this week.
先週の進捗(ZHOU)
- Finished the testing of EfficientDet's GUI. Thank Professor Zhang for the instruction of camera parameter's setting.
- Finished the PPT of presentation in ICIAE2021.
- Moved to the new house, thanks for everyone's help.
先週の進捗(ZHOU)
- Re-modified the paper of ICIAE, and upload it to gitlab.
- Because different objects reflect different light, the background with different object will be different. Inspired by this, for a combination, one of these image will be considered as background, and the other objects will be extracted and put in the background. Therefore, EfficientDet can … Continue Reading ››
先週の進捗(ZHOU)
Tried to modify the datasets of vending supermarket by interpolating between object and background, but failed. I will try some other methods.
先週の進捗(ZHOU)
Re-modify the datasets of vending supermarket, remove the effect of object's background. The object is extracted, instead of both the background and object.
Validating and testing EfficientDet, the result is show in nextcloud(https://mountain.elcs.kyutech.ac.jp:8443/apps/files/?dir=/%E5%80%8B%E4%BA%BA%E7%94%A8_%E6%95%99%E5%93%A1/%E5%BC%B5/QT_zhou_dataset&fileid=1131808), the result on testing set is not good. I think may be a light need … Continue Reading ››