-
I use ABCNet to train my own dataset. The dataset sample just like:
![1](https://user-images.githubusercontent.com/15357846/99898974-c4bdc680-2ce0-11eb-8628-03e70258047d.png)
Use `abcnet_custom_data…
-
Thanks for your excellent works!
I have some questions for condinst, have you tried use conditional conv to predict regressions for bbox? I am wondering what's the performance.
-
Hi, Zhi
I implement CondInst in my own code and get 38.9 box mAP and 4.3 segm mAP.
When debugging, I find that the GT assignment of the proposal generator is not consistent with the origin FCOS.…
-
Hi,
I just run the default ConInst with condinst_MS_R_50_1x config, 8 GPUs, and 16 batch sizes in total as described in this repo. However, I get poor performance.
![image](https://user-images.gi…
-
Hi, thank you for your nice work!
I'm wondering whether you have contucted the experiments where the **normal conv layers** are used instead of conditional convolution in the mask head, just like T…
-
I'm interested in this work, but I don’t know much about the task of instance segmentation, and I don't have a very clear understanding of the following questions:
1. How to determine the actual numb…
-
Hi~ @Epiphqny
Thanks for your code ~ I run your code and have some questions about your implementation.
1. The mask tower part is not consist with the paper.
https://github.com/Epiphqny/CondInst…
-
Hi~
Is `batchsize_per_GPU = 2` enough for CondInst? I found the default norm type of the `mask_branch` is batchnorm, which is same as BlendMask, but for BlendMask, the `batchsize_per_GPU = 4`.
-
### **Abstract**
  针对目前大多数的多目标跟踪方法的检测与跟踪关联性较弱,在处理上相对独立的问题,本文提出了一个联合检测于跟踪的模型,利用跟踪的线索来端到端的辅助检测。通过cost volume来推测跟踪的offset,并用来聚合之前的特征来辅助提升检测与分割的性能。在MOTChallenge(2D)、nuScenes(3D)、MOTS以及Youtube-V…
-
Hi, when I run condinst, I meet the question as follow,
-- Process 1 terminated with the following error:
Traceback (most recent call last):
File "/home/l547/anaconda3/envs/detectron2/lib/python3…