-
模型:cascade_rcnn_mobilenetv3_fpn_640.nb
手机:骁龙625
使用PaddleLite-andriod-demo里的object_detection_demo在Andriod Studio上部署到手机上。使用里面已经有的模型可以部署到手机成功运行,替换成cascade_rcnn_mobilenetv3_fpn_640.nb后,可以部署到手机,也可以load m…
-
I'm using the `DefaultPredictor` with the `Misc/cascade_mask_rcnn_X_152_32x8d_FPN_IN5k_gn_dconv.yaml` configuration.
Given its prediction output for one image, I'd like to know the recommended appr…
-
I found that some weights of GCNET are wrong and do not match the corresponding config.
For example, the last row of the table.
Config: https://github.com/open-mmlab/mmdetection/blob/master/con…
-
I used detectron2 Cascasde-mask-rcnn to train myself dataset and got mAP about 0.75 for BBox and segm at day 5/16/2020.
Today I use the same trained model to run test procedure again but got mAP onl…
-
## 📚 Documentation
I used the coco dataset to train centernet2, but got a error:
```
Traceback (most recent call last):
File "./train_net.py", line 245, in
args=(args,),
File "/dat…
Lg955 updated
3 years ago
-
Hello, I've converted Cascade-RCNN-R50-FPN-coco to ONNX format but when I perform inference with it by onnx-runtime, the final mAP drops from 40.3 to 39.7. Same thing happens with Retinanet. So may I …
-
Hello, sir. I use cascade rcnn to train my dataset and use dist_train.sh with two gpu, but I found that the memory used in two gpus is different,
![image](https://user-images.githubusercontent.com/4…
-
- 版本、环境信息:
1)PaddlePaddle版本:1.8.0
2)系统环境|GPU:aistudio上,v100
- 训练信息
1)单卡
2)16G
3)[operator < read > error]
- 复现信息:使用官方PaddleDetection-release-0.3,配置文件cascade_mask_rcnn_dcnv…
-
在train的时候use_padded_im_info:false,eval和test的时候为true。在operator.py看到源码信息为if self.use_padded_im_info: data['im_info'][:2] = max_shape[1:3]。水平问题看不太懂源码,能麻烦解释一下为什么训练和验证时分别设置为false和true吗,代表什么意思?谢谢
-
您好, 感谢回复
使用版本 release/2.0-beta
在cpp推理时,batch不论是否大于1时,模型推理结果都为 n x 6的结果, 这是包含了 nms 后的结果,而且模型输出只有这一个tensor,因此目前认为无法区分这n个数据无法和batch个图片做对应关系
当然,知道n x 6经过 nms 后是按照 score 排序了的, 但是如果其中有一张图片一个框都没有,这时候就无法知道…