{calayzhou,ybwang,lvtao,lyq,linsen chen}@smail.nju.edu.cn, {shenqiu,caoxun}@nju.edu.cn Abstract We endeavor on a rarely explored task named Insubstan- tial Object Detection (IOD), which aims to localize the ob- ject with following charac...
Dense-captured images of a scene are sometimes not available, so we also study the sparse-view scenario here. The following methods are included: NeuS [58] with sparse-view input; MonoSDF [72], which takes in geometry cues estimated by pre-trained ...
However, most of the following pose and shape estimation methods do not include such a confidence threshold.3 Instead they assume M to be given as stated in Problem 1. To still follow the same evaluation protocol as [3], all other methods rely on the same, suboptimal Mask R-CNN ...
Saliency maps were also calculated for an ideal observer39,42. The ideal observer evaluated the importance of different object parts based on the pixel space information they provided for recognition. In other words, the ideal observer, and the following classification algorithm, used original masked ...
I closed the development of the .z3m export module (to create the .z3m file without using the original .z3m file) based on the following problem: The vertex bone reference values (range: 0-255) are much more greater than the number of used bones (usually: about 26). I uploaded th...
We found that algorithms using these modalities as input were difficult to fit into the scope of this work, for the following reasons: Besides CosyPose [5], we did not find any multi-view approaches that fit the scope given by our use case. Point-cloud-based object detectors are very ...
Outliers and errors are inevitable in pairwise registrations, and the transformation relationships in 𝑻𝑔𝑙𝑜𝑏𝑎𝑙Tglobal are not always accurate. 𝑻𝑔𝑙𝑜𝑏𝑎𝑙Tglobal can be decomposed into the following equation: 𝑻𝑔𝑙𝑜𝑏𝑎𝑙=𝑻̂𝑔𝑙𝑜𝑏𝑎𝑙...