LWJ312 / ION

This is the project of Instance Object Navigation for ACM MultiMedia 21
7 stars 3 forks source link

some problems about"Note that if multiple instances of the same category are detected, the instances selection will be taken to select the best fit."in paper #1

Open Rh-Dang opened 2 years ago

Rh-Dang commented 2 years ago

Dear Weijie LI if the instances are not in the same category, Is it still useful to choose color and material similar instance to the target object

LWJ312 commented 2 years ago

Thanks for your attention to our work. The "Instance Selection" only works for those instances (of the same category) in the current view (see Paragraph 1 and Paragraph 3 in Section 4.1). So for what you said "if the instances are not in the same category" , this means that in that category there is only one instance, so there is no choice to select "other instances", therefore "choose color and material similar instance to the target object" doesn't work for this situation at all. For example, if the agent sees "two different cups, a bread and a bowl", our work will select one cup between these two cups, and embed this selected cup as well as the bread and bowl into the IRG. Hope that can answer your question.

Rh-Dang commented 2 years ago

你好,很感谢您耐心的回答!!我英语不好,既然我们都是中国人,我们就用中文交流吧: 我想您没有理解我的问题。我是对您论文中的(1)公式产生的疑惑。根据我的理解您的论文中(1)公式得作用是在视野中有多个相同类别物体时选择其中一个。但是(1)公式的选择方式是计算与目标物体颜色和材质最相近的那个,然后选择它。比如我们要导航到一个苹果,而此时刻的视野中有“两个杯子,一个面包,一个碗”,那么计算杯子和导航目标(苹果)的材质和颜色的相似度就是让人无法理解的。 也有可能是我没有充分理解公式(1),希望您能解答我的疑惑。万分感谢!!

Thanks for your attention to our work. The "Instance Selection" only works for those instances (of the same category) in the current view (see Paragraph 1 and Paragraph 3 in Section 4.1). So for what you said "if the instances are not in the same category" , this means that in that category there is only one instance, so there is no choice to select "other instances", therefore "choose color and material similar instance to the target object" doesn't work for this situation at all. For example, if the agent sees "two different cups, a bread and a bowl", our work will select one cup between these two cups, and embed this selected cup as well as the bread and bowl into the IRG. Hope that can answer your question.