I guess there is something wrong about the released network or at least the preprocessing code. I tried to use prediction-with-pretrained example but the results are mistaken.

I also realized that output layer has 1008 nodes where as the label txt has 1001 classes

This repository is over its data quota. Purchase more data packs to restore access.

It seems that there is no synset.txt file for VGG models at http://www.mxnet.io/models/imagenet/vgg/

I'm wondering where can I found the corresponding labels?

Hello, the following link:
http://data.dmlc.ml/mxnet/models/imagenet/inception-21k.tar.gz
mentioned in:
https://github.com/dmlc/mxnet-model-gallery/blob/master/imagenet-21k-inception.md
no longer works and http://data.dmlc.ml redirects to anther site where the model does not seem to be accessible.
Where can I access the model?
Thanks.
Georges.

It's confusing that the zip files can't be decompressed directly. Most users may give up using your models if they fail to figure out that they have to use git lfs.

Hi,

Apart from this imagenet Model trained on all the 21k synsets, are there any other models that have done the same ?
for example ResNet with 21k+ classes trained ?

I can not find pretrained model of inception-v4 and inception-resnet-v2 here. I think it is proper to add them here.
Does anyone have one?

Hi

When I run classification on an image, i am able to extract the top N predicted tags. I can also extract the 1000 or 21000+ length probability array for each class.

Now, for example I have an image of a house with 2 doors. The Classifier can predict door with some probability. I wanted to know if i can tap into some layer of the model to find out how many doors were there ? I am assuming that the classifier will update the probability of the door , if it encounters it for he second time. So, is there a way I can count the number of such tags or such synsets that it has predicted by the classifier ?

The pre-trained imagenet model is super useful. A model pre-trained on wikipedia (or another large text corpus) would be super useful too!

I noticed that there are different image pre-processing pipelines for different models.

For Inception-BN, the images should be normalized with the mean_224.nd:

mean_img = mx.nd.load("Inception/mean_224.nd")["mean_img"]
normed_img = sample - mean_img.asnumpy()

For Inception-V3, the images should be:

normed_img = sample - 128.
normed_img /= 128.

For Inception-BN-21k, the images should be:

normed_img = sample - 117.

Am I correct with the above settings?

I find 1000 classes id not include in 21000 classes id, so where can i get correspondence.txt?

Hi,

I am trying to download the above model using the following:

wget http://data.dmlc.ml/mxnet/models/imagenet/inception-v3.tar.gz

but the download is not a success. I would be much thankful for your kind support in this regard.

Thank You &
Warm Regards,
Piyumal

I want to download the new inception-v3 file, but find the lfs data quota is exceeded. Is there any alternative way to access the model zip file?

Here is what I've got after running lfs pull:

(0 of 3 files) 0 B / 244.38 MB
This repository is over its data quota. Purchase more data packs to restore access.
Docs: https://help.github.com/articles/purchasing-additional-storage-and-bandwidth-for-an-organization/

Hi,

I have been trying to download the Inception-V3 model from the given website from the past couple of days but, the link seems to be down. Has the file been shifted?

Thanks,
Daksh

In https://github.com/dmlc/mxnet-model-gallery/tree/master/imagenet-21k-inception:

This network runs roughtly 2 times ...

should be

This network runs roughly 2 times ...

Are given test values relying on single view predictions or multi-view values?

I can connect the host,but can't download. 404 not found error. It seems the file not exists.

#Hi,

I got an RuntimeError: prob_label is not presented when I run [http://mxnet.io/tutorials/python/predict_imagenet.html](Example : Predict with pretrained model) with pretrained model https://github.com/dmlc/mxnet-model-gallery/blob/master/imagenet-1k-nin.md.

first , I load pretrained model follow the tutorial:

import mxnet as mx
sym, arg_params, aux_params = mx.model.load_checkpoint('nin', 0)

I found 'prob_label' in sym.list_arguments() but it is not exists in 'arg_params'

when I create a model for this model on GPU 0 like this:

mod = mx.mod.Module(symbol=sym, context=mx.gpu())
mod.bind(for_training=False, data_shapes=[('data', (1,3,224,224))])
mod.set_params(arg_params, aux_params)

the error occurs at mod.set_params(arg_params, aux_params) ,

Traceback (most recent call last):
  File "loadmodel.py", line 7, in <module>
    mod.set_params(arg_params, aux_params,allow_missing=False)
  File "/usr/local/lib/python2.7/dist-packages/mxnet-0.7.0-py2.7.egg/mxnet/module/base_module.py", line 483, in set_params
    allow_missing=allow_missing, force_init=force_init)
  File "/usr/local/lib/python2.7/dist-packages/mxnet-0.7.0-py2.7.egg/mxnet/module/module.py", line 198, in init_params
    _impl(name, arr, arg_params)
  File "/usr/local/lib/python2.7/dist-packages/mxnet-0.7.0-py2.7.egg/mxnet/module/module.py", line 191, in _impl
    raise RuntimeError("%s is not presented" % name)
RuntimeError: prob_label is not presented


how can I fix this?

Is anyone also experiencing this?

Do you have a pretrained model for predicting NSFW images using mxnet?

thanks.

Hi all!
I would be interested in classify just few of all the labels (seashore, lakeshore and alp). How could I go through this, maybe modifying one of the already existing pre-trained model? Thanks in advance!

My code as follows:

122 cv::Mat im_ori = cv::imread(image_file, 1);
123 /*
124 * preprocess image as Inception_v3 required
125 * crop -> resize -> normlize (-mean)/std
126 /
127 int short_edge;
128 if(im_ori.rows > im_ori.cols)
129 {
130 short_edge = im_ori.cols;
131 }
132 else
133 {
134 short_edge = im_ori.rows;
135 }
136 //printf("image size, row = %d, col = %d, short_edge = %d\n", im_ori.rows, im_ori.cols, short_edge);
137 int yMin = (im_ori.rows - short_edge) / 2;
138 int xMin = (im_ori.cols - short_edge) / 2;
139 int xMax = xMin + short_edge;
140 int yMax = yMin + short_edge;
141 cv::Mat croppedImg;
142 im_ori(cv::Rect(xMin,yMin,xMax,yMax)).copyTo(croppedImg);
143 //cv::imwrite("ori.jpg", im_ori);
144 //cv::imwrite("crop.jpg", croppedImg);
145
146 cv::Mat im;
147 resize(croppedImg, im, resize_size);
148 //cv::imwrite("resize.jpg", im);
149 int size = im.rows * im.cols;
150 mx_float ptr_image_r = image_data;
151 for(int i = 0; i < im.rows; i++)
152 {
153 uchar* data = im.ptr(i);
154 for(int j = 0; j < im.cols; j++)
155 {
156 mx_float r = (data[j] * 256 - 128) / 128.0;
157 *ptr_image_r++ = r;
158 }
159 }

I read preprocessing.py, it first crop, then resize, at last normalization.

Are my codes wrong?

The Inception-BN Network in mxnet-model-gallery needs to be updated:

1. The inception-bn.tar.gz contains the old model that doesn't work any more.
2. imagenet-1k-inception-bn.md doesn't contain information about the file 'mean_224.nd'(in the tar.gz) that the inception model depends on.

The link http://data.dmlc.ml/mxnet/models/imagenet/squeezenet/squeezenet_v1.0.tar.gz for squeezenet model is not working. Please upload the squeezenet model

Hi I am trying to use Inception V3 for my image classification project. But while trying to load the model, its giving me download error as below

model=InceptionV3()
Downloading data from https://github.com/fchollet/deep-learning-models/releases/download/v0.5/inception_v3_weights_tf_dim_ordering_tf_kernels.h5

Exception: URL fetch failure on https://github.com/fchollet/deep-learning-models/releases/download/v0.5/inception_v3_weights_tf_dim_ordering_tf_kernels.h5 : None -- [WinError 10060] A connection attempt failed because the connected party did not properly respond after a period of time, or established connection failed because connected host has failed to respond

Howevr I have download the "inception_v3_weights_tf_dim_ordering_tf_kernels.h5" file, but I don't know how to use this file.

Please help me solve either of my problem.

Thanks

Hi there,

I appreciate the pretrained model shared on links. However, I want to try adjustments on these models, so surgery on symbol.json does not seem like the proper way. where could I find the source code for symbol.json?

Could you please help in creating features (output at the end of Convolution layers and before MLP) using pre-trained network for smaller images?

I am working with images of 11*11 size and have fairly large dataset of about 300K images.

Following example in following link: http://mxnet.io/tutorials/r/classifyRealImageWithPretrainedModel.html

Problem is here net requires images of size 224*224.

Hi,

I am trying to convert the Inception21k model to PyTorch. I used the MMdnn tool to do the conversion; however, I am only seeing 65% top1 accuracy on the validation set (below the reported 68%).

What was the image preprocessing used to train the network? Currently I am doing:

1. Resize shortest size of image to 256
2. Center crop to 224
3. Multiply by 255 to scale image input range to [0, 255]
4. Subtract mean of 117 from the image

Is this correct?

It seems that the URL for VGG model is not correct in the description.

The model is located at

http://www.mxnet.io/models/imagenet/vgg/

instead of

http://data.dmlc.ml/mxnet/models/imagenet/vgg

Hi,
Can you detail how to convert the inception-v3 model from tensorflow to mxnet format.
Regards,
Ramya