wanglimin / mrcnn-scene-recognition Goto Github PK
View Code? Open in Web Editor NEWMR-CNNs for Large-Scale Scene Recognition
MR-CNNs for Large-Scale Scene Recognition
Hi wanglimin,
Thank you for sharing your source about scene classification.
But it seems that the pre-trainded models can not be downloaded, when run the script of get_init_models.sh
.
您好,我自己写了一个python脚本
MODEL_FILE = 'models/standard_train/256_inception2_deploy.prototxt';
PRETRAINED = 'models/places2_standard_256_inception2_v5.caffemodel';
caffe_root="/home/xxx/work/deeplearning-tools/caffe/caffe-v0.9999"
import caffe
#GPU模式
caffe.set_mode_gpu()
#定义使用的神经网络模垿
net = caffe.Classifier(MODEL_FILE, PRETRAINED,
mean=np.load(os.path.join(caffe_root,'python/caffe/imagenet/ilsvrc_2012_mean.npy')).mean(1).mean(1),
channel_swap=(2,1,0),
raw_scale=255,
)
## execute
sample = caffe.io.load_image("Places365_val_00000041.jpg")
#预测图片类别
prediction = net.predict([sample])
来测试您的模型,报错如下:
Message type "caffe.BatchNormParameter" has no field named "engine"
我google了一下这个错误,没有找到答案,能给点建议吗?
谢谢!
Can you tell me how to run this project?Is there a description document?
Hello Mr.Wang!
I am trying to test your demo.But I can't find the data file on the internet ,which is necessary in your code.It would be wonderful if you offer me this file! Thanks very much!
the readme show that you have released the mode and test code ,but not find, where can i get them?
Hi,
I tried to download the pre-trained knowledge models and reference models using the scripts get_init_models.sh
and get_reference_models.sh
but all the model files are missing (404 Not Found Error).
Could you please let me know the correct url of these models so that I can download them? Thanks so much!
Hello,
I am unable to test the code by using the instructions as you described here. The error as below;
Attempt to reference field of non-structure array.
Error in test_places2 (line 16)
file_list = tmp.textdata;
I have another question;
What should be the following files are;
file_name = 'places2/val_result.mat';
val_file = 'places365_val.txt';
Thanks in advanced.
I appreciate your work on the paper "Knowledge Guided Disambiguation for Large-Scale Scene Classification with Multi-Resolution CNNs"。 I could not find the trained model you mentioned! Could you please help me, thank you very much。
王大神,你好!
我想问一下,论文中提到lsun的网络是在imagenet/places2上微调的,但是lsun只有10个场景,numout应该是10啊?为什么代码里的numout是365呢?
Hi,
Many thanks for sharing. Truly appreciate your amazing works.
I was going to apply your pre-trained model to my own dataset; however, the permutation of input images' height and width makes me confused:
At line 33 of matlab/test_places2.m and line 40 of matlab/test_mit67.m, you performed
im_data = permute(im_data, [2, 1, 3]); % permute width and height
So, you rearrange the shape of the input image from (height, width, depth) to (width, height, depth), where depth=3 reflecting 'BGR' color channels; am I right?
Could you please explain a little bit on this operation? Is it a convention to do this, or there are other deeper reasons?
BTW: I'm using OpenCV with Python instead of Matlab. In Python, the image is loaded of shape (height, width, depth='bgr'); and I noticed that for Caffe, the conventional blob dimensions for batches of image data are number N x channel K x height H x width W. I'm not sure if there's any relation with the permutation.
Thanks for your time.
I found that in the train_val the loss_weight of kd-loss is set to 0.25. But in the paper the lambda is set to 0.5(The parameter of = 0.5 is the best choice for scene network disambiguation for both normal BN-Inception and deeper BN-Inception architectures. ). Is there any wrong?
When I scan your code file(modles/kd_train/*_train_val.prototxt) yesterday,I found an unkonw layer type named SoftmaxWithCrossEntropyLoss which output 'kd-loss'.I have checked a lot of information include caffe wrote by yjxiong,but I still can't find some information about it.
Maybe I can replace it with SigmoidCrossEntropyLossLayer?
Dear author:
I am sorry to trouble you, but I just want to know the speed of this new method. Please tell me the information of GPU that you use. And how many pictures can you recognize during 1 second?
I am looking forward to getting your reply, thank you! @wanglimin
电脑只有一个GPU,想直接用官方caffe,请问您可以吗?
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google ❤️ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.