lyxok1 / tiny-dsod Goto Github PK

hello sir,
the caffe which gets build here is unable to link the modules of caffe. According to the below thread
[https://github.com/BVLC/caffe/issues/4622](url) the error I think error is from your side.
Can you help in resolving the issue.
Here is the photp of attached error.
File "", line 1, in
File "/home/root1/Desktop/PRAVESH/TINY-DSOD/python/caffe/init.py", line 1, in
from .pycaffe import Net, SGDSolver, NesterovSolver, AdaGradSolver, RMSPropSolver, AdaDeltaSolver, AdamSolver
File "/home/root1/Desktop/PRAVESH/TINY-DSOD/python/caffe/pycaffe.py", line 13, in
from ._caffe import Net, SGDSolver, NesterovSolver, AdaGradSolver,
ImportError: dynamic module does not define init function (init_caffe)

Hello! i was just reading the paper and i would like to give it a try! but i am not able to download the pre trained model from baidu, could you please upload it to another platform? thanks!

math_functions.cu:79] Check failed: error == cudaSuccess (77 vs. 0) an illegal memory access was encountered.
i use the following command to train the model
caffe.exe train -solver="solver.prototxt" -weights="Tiny-DSOD.caffemodel" -gpu 0
i get the error above。
I get the same error even when i do not use the pretrain model。

The pre-trained weights download link links to a 'baidu download' server in Chinese. I'm not able to read Chinese and don't know how to proceed for downloading. Could you please be so kind to provide a direct link or a link to an English download server (maybe upload it to github, it's only 3.8MB)?

Thanks

Oh....., is there anyone who has the same problem? I have been trapped in this porblem for a long time.

I get accuracy of 60% on Pascal VOC. Anyone has gotten near 70%?

I've properly followed your readme steps and while trying to build the detector, I've encountered
some recurring problems regarding the caffe framework.

e.g.:
src/caffe/net.cpp:29:1: error: prototype for ‘caffe::Net::Net(const caffe::NetParameter&, const caffe::Net)’ does not match any in class ‘caffe::Net’
Net::Net(const NetParameter& param, const Net root_net)
^
In file included from src/caffe/net.cpp:11:0:
/usr/local/include/caffe/net.hpp:339:1: error: candidates are: caffe::Net::Net(const caffe::Net&)
DISABLE_COPY_AND_ASSIGN(Net);
^
In file included from src/caffe/net.cpp:13:0:
/usr/local/include/caffe/net.hpp:27:12: error: caffe::Net::Net(const string&, caffe::Phase, int, const std::vector<std::__cxx11::basic_string >)
explicit Net(const string& param_file, Phase phase,
^
/usr/local/include/caffe/net.hpp:26:12: error: caffe::Net::Net(const caffe::NetParameter&)
explicit Net(const NetParameter& param);
^
src/caffe/net.cpp:35:1: error: prototype for ‘caffe::Net::Net(const string&, caffe::Phase, int, const std::vector<std::__cxx11::basic_string >, const caffe::Net)’ does not match any in class ‘caffe::Net’
Net::Net(const string& param_file, Phase phase,
^
In file included from src/caffe/net.cpp:11:0:
/usr/local/include/caffe/net.hpp:339:1: error: candidates are: caffe::Net::Net(const caffe::Net&)
DISABLE_COPY_AND_ASSIGN(Net);
^
In file included from src/caffe/net.cpp:13:0:
/usr/local/include/caffe/net.hpp:27:12: error: caffe::Net::Net(const string&, caffe::Phase, int, const std::vector<std::__cxx11::basic_string >)
explicit Net(const string& param_file, Phase phase,
^
/usr/local/include/caffe/net.hpp:26:12: error: caffe::Net::Net(const caffe::NetParameter&)

I do believe I've followed your steps correctly, hence I have some questions:

• Is this a known problem?
• What version of caffe are you guys using?
• Am I perhaps missing out on some libraries that aren't explicitly mentioned within your readme?

thanks in advance!

How do you calculate your model's FLOPs? I wonder which parts of the model do you count?

Traceback (most recent call last):
File "examples/DCOD/DCOD_pascal.py", line 409, in
DCOD300_Body(net, from_layer='data',growth_rate=32, use_reverse=use_reverse, use_objectness=use_objectness)
NameError: name 'DCOD300_Body' is not defined

Hello,
OpenCV allows to read/process DNN models using cv::dnn::readNetFromCaffe. However some layers are not implemented yet in particular "Upsample" and "conv_dw" layers. Leading to the following error during the forward pass since those layers are not implemented

Error: Unspecified error (Can't create layer "Upsample1" of type "Upsample") in cv::dnn::LayerData::getLayerInstance, file \opencv\modules\dnn\src\dnn.cpp, line 518

Does anyone have managed to add those layers to the DNN modules of OPENCV?

Thanks !!!

Thanks for sharing you code. I complied this version of caffe, but when i tried to train the model, the training speed is so slow even though i used two Titanx gpu. I cannot figure out why is that. Could you help me out? very thanks

Thanks for your work, I works on tx2.But I think the inference time is not normal，MobileNetV1-ssd can run 38pfs on Jetson TX2，Tiny-DSOD only 14fps.
Is it wrong with my operation?
Thanks

the video detection module of Tiny-DSOD fails first gives below error:

Traceback (most recent call last):
File "examples/DCOD/video_detection_demo.py", line 198, in
fps = (batch_size*len(batch_list))/total_time
ZeroDivisionError: float division by zero

When the total_time value is set equal to time of video I get below error:
I0616 17:45:29.350755 10882 net.cpp:781] Ignoring source layer mbox_loss

(python:10882): GLib-GObject-CRITICAL **: g_object_set: assertion 'G_IS_OBJECT (object)' failed
finish reading videos...
Traceback (most recent call last):
File "examples/DCOD/iith_video_detection_demo.py", line 205, in
output_size = (output[0].shape[0], output[0].shape[1])
IndexError: list index out of range

Can you help how to proceed for video detection.

首先感谢无私分享，我有个疑问：在你的upsample_layer.cpp的使用的是双线性插值，正常的双线性插值对齐应该写成onst Dtype b1_x = (x+0.5)/w_ratio_ - 0.5，而你的65行写成onst Dtype b1_x = (x+1)/w_ratio_ - 1.0;使用的1.0会导致对齐偏向图像右下角，也就是特征图的左上角值不会参与使用，这样做的目的是什么，还是我理解有问题？

After make -j8 i get following error:-

Makefile:619: recipe for target '.build_release/tools/upgrade_net_proto_text.bin' failed
make: *** [.build_release/tools/upgrade_net_proto_text.bin] Error 1

What all changes do we need to do in the config file ?
Any help would be appreciated

It would be much helpful if you share it... Thanks very much!

Hi， guys，
I used tiny-dsod to train a idcard-detection model. It perferms well. But for pictures without idcard it falsely detected some "idcard".

The training data cotained pictures each has at least one idcard inside.

we tried to train the model on pure background dataset which with no idcard rect. the loss is always zero. but we test the model, it also falsely detected some "idcard".

Can you give me any sugguestion？

finish processing batch 74
finish processing batch 75
finish processing batch 76
Traceback (most recent call last):
File "examples/DCOD/video_detection_demo.py", line 198, in
fps = (batch_size*len(batch_list))/total_time
ZeroDivisionError: float division by zero

Can you share the pretrained model on Kitti please?
Thank you!

could you give us more information on how to modify the Makefile ?
thanks

Tiny-DSOD$ make
CXX src/caffe/common.cpp
src/caffe/common.cpp: In constructor ‘caffe::Caffe::Caffe()’:
src/caffe/common.cpp:109:42: error: class ‘caffe::Caffe’ does not have any field named ‘root_solver_’
mode_(Caffe::CPU), solver_count_(1), root_solver_(true) {
^
Makefile:576: recipe for target '.build_release/src/caffe/common.o' failed
make: *** [.build_release/src/caffe/common.o] Error 1

Could you share your pre-trained weights on COCO dataset

Hi,

When I try to run the solver to perform the training, Caffe didn't recognize "ConvolutionDepthwise" layer type.

Check failed: registry.count(type) == 1 (0 vs. 1) Unknown layer type: ConvolutionDepthwise

Thanks

错误如下

Check failed: target_blobs.size() == source_layer.blobs_size() (1 vs. 2) Incompatible number of blobs for layer Convolution1

我个人认为是bias设置的问题，权重文件里有的层有bias但是deploy设置了false，我一个一个修改后，检测结果一直为空

I found this moel can run successfully on image detection.But when it comes to video detection, it failed.
Whole memory is used (8G).

thanks for your work, and I want to know what is the result of this work on small object detection?