LMM

Have some questions about the structure?

1

pspnet_unet Auxiliary_head The auxiliary decoding head uses fcn, but this will cause the structure parameters to be too large. Can the auxiliary decoding head use a lightweight method, such as...

mobilenet_v3 The dataset used can split the dataset for voc, if so how to modify it.

3

Because I see that most of the datasets are cityscapes, but there are relatively few data sets using voc_aug. If I use voc in these networks, I need to modify...

Why is the size of the model file generated by the training file different from the actual file size?

5

Hello! I have a question here, that is, the size of the file generated by the training is twice as large as the size of the model provided by your...

[General Issue] bankbone修改之后，似乎出现了问题，分割图像时出现了，像素点错乱

3

pspnet 修改如下： model: type: PSPNet backbone: type: MobileNetV3_large_x1_0 pretrained: https://paddleseg.bj.bcebos.com/dygraph/backbone/mobilenetv3_large_x1_0_ssld.tar.gz backbone_indices: [1, 3] pp_out_channels: 1024 bin_sizes: [1, 2, 3, 6] enable_auxiliary_loss: True align_corners: False pretrained: null 但是效果出现了问题，出现了像素点分割散乱，是不是配置存在没有对上的问题，但是代码它能够正常训练，能看一下是哪里出现了问题嘛

[Bug] paddle2onnx 导出出现了问题

1

欢迎您反馈PaddleSeg使用问题，辛苦您提供以下信息，方便我们快速定位和解决问题： 1. PaddleSeg版本：PaddleSeg==2.6 2. PaddlePaddle版本：（如PaddlePaddle 2.1.0） 3. 操作系统信息：Linux 4. Python版本号：Python3.6.13 5. CUDA/cuDNN版本：CUDA11.2 Traceback (most recent call last): File "/home/dell/anaconda3/envs/d2l/lib/python3.6/site-packages/paddle2onnx/op_mapper/op_mapper.py", line 119, in mapping mapper_func(graph, node, **kw) File "/home/dell/anaconda3/envs/d2l/lib/python3.6/site-packages/paddle2onnx/op_mapper/nn.py", line 186,...

[Bug] ESPNet网络训练TypeError

4

espnet网络训练过程中出现了 TypeError: 'tuple' object does not support item assignment。 日志如下： 022-07-27 11:27:25 [INFO] ------------Environment Information------------- platform: Linux-5.4.0-121-generic-x86_64-with-debian-buster-sid Python: 3.6.13 |Anaconda, Inc.| (default, Jun 4 2021, 14:25:59) [GCC 7.5.0] Paddle compiled...

如何为PSPNet修改bankbone

2

我想将pspnet的主干bankbone改为主干为mobnetv3网络，那么这样我是否需要修改源码，如果不需要的话，那么需要更改那个配置，谢谢！期待您的回复？

转化出来的onnx结构不是官方通用的，一些算子有没有什么办法解决

22

转化出来的结构推断好像不是官方通用的输出，如resize算子转化出来不是官方通用的，这个有没有什么办法解决嘛？  再使用onnxsim优化其结构时出现了一些卷积操作推断不出来，这个是不是哪里存在问题，能帮我解答一下疑惑嘛。  我是用于模型推理部署

paddle2onnx导出模型失败

7

PaddleSeg版本：PaddleSeg==2.6 PaddlePaddle版本：PaddlePaddle 2.1.0 操作系统信息：Linux Python版本号：Python3.6.13 CUDA/cuDNN版本：CUDA11.2 Traceback (most recent call last): File "/home/dell/anaconda3/envs/d2l/lib/python3.6/site-packages/paddle2onnx/op_mapper/op_mapper.py", line 119, in mapping mapper_func(graph, node, **kw) File "/home/dell/anaconda3/envs/d2l/lib/python3.6/site-packages/paddle2onnx/op_mapper/nn.py", line 186, in opset_1 node.input_shape('X', 0), node.output_shape('Out', 0)))...

How to choose loss?

3

Only one category is segmented, but there are too many pixels in the background category, and there are too few pixels in the category to be segmented. How to choose...