CenterNet-CondInst icon indicating copy to clipboard operation
CenterNet-CondInst copied to clipboard

Instance Segmentation based on CenterNet and CondInst

Results 12 CenterNet-CondInst issues
Sort by recently updated
recently updated
newest added

instance_masks = np.zeros((num_objs if num_objs > 0 else 1, output_h, output_w), dtype=np.float32) 这里instance_masks的大小是和objs数相关的,那一个batch中每张图可能不一样,这个在组batch的时候如何处理?

``` class COCO(data.Dataset): num_classes = 80 default_resolution = [512, 512] mean = np.array([0.40789654, 0.44719302, 0.47026115], dtype=np.float32).reshape(1, 1, 3) std = np.array([0.28863828, 0.27408164, 0.27809835], dtype=np.float32).reshape(1, 1, 3) ``` https://github.com/CaoWGG/CenterNet-CondInst/blob/1755e1752301e08363472b5e45c90723111877e9/src/lib/datasets/dataset/coco.py/L13-L19

非常感谢您分享的代码。 由于我暂时还没有配置成功代码,所以只能干看代码,没法调试。 有一个问题想请教一下,就是原始的CenterNet的输出有三个分支,分别是 heatmap (W\*H\*C),offset (W\*H\*2)和size (W\*H\*2),然后你这里加了一个seg_feat,这个分支是怎么加的,能介绍一下吗?能否告知是在代码的哪一处?这里的seg_feat它的size是什么样子的?怎么为每个中心点分配一个mask?难道与offset和size一样,预测一个 W\*H\*W\*H的seg_feat? 此外,代码中关于dice loss的计算,我也不是很明白: ``` def forward(self, seg_feat, conv_weight, mask,ind, target): mask_loss=0. batch_size = seg_feat.size(0) weight = _tranpose_and_gather_feat(conv_weight, ind) h,w = seg_feat.size(-2),seg_feat.size(-1) x,y =...

Hello, I have a custom dataset annotated in Coco format with coco-annotator. How can I train the data using the model

https://github.com/CaoWGG/Mask-CenterNet/blob/0b40993135acafa3e5a85a5311a49907225c9530/lib/datasets/sample/ctseg.py#L134

我跑了一下2017的数据集,貌似准确率不太高,问一下可以出个detectron2版吗,源码绕来绕去不好迁移

使用demo.py进行训练效果展示的时候只画出了框,但是没有mask

the figure in the readme contains FPN. Do you use FPN in the backbone?