Can we use caffe for other image formats like .pgm?

[Sateesh Babu G]

Can we use caffe for other image formats like .pgm?

[Ashwani Rai]

Yes. I've used it with .ppm

[Sateesh Babu G]

@ Ashwani Rai Could you please describe the procedure to use caffe for .ppm images for classification ?

[Ashwani Rai]

Nothing special, same as jpeg example.

[Sateesh Babu G]

@ Ashwani Rai, How to use caffe for gray scale images like .pgm format?

[Michael Wilber]

Caffe's convert_imageset tool uses OpenCV to read images, and the list of supported image formats is documented here: http://docs.opencv.org/3.0-beta/modules/imgcodecs/doc/reading_and_writing_images.html#Mat imread(const String& filename, int flags) 

• Windows bitmaps - *.bmp, *.dib (always supported)
• JPEG files - *.jpeg, *.jpg, *.jpe (see the Notes section)
• JPEG 2000 files - *.jp2 (see the Notes section)
• Portable Network Graphics - *.png (see the Notes section)
• WebP - *.webp (see the Notes section)
• Portable image format - *.pbm, *.pgm, *.ppm (always supported)
• Sun rasters - *.sr, *.ras (always supported)
• TIFF files - *.tiff, *.tif (see the Notes section)

If it doesn't work, you can also use the special "ImageMagick" preprocessing layer :) Run "convert imagename.ppm imagename.png" with ImageMagick installed before feeding them through converge_imageset.

[Sateesh Babu G]

@ Michael Wilber, Thanks for your confirmation. I am new to Caffe, Could you please list procedure to train and classify gray scale images using Caffe.

[Ashwani Rai]

@Sateesh Just follow the example provided on classification on the caffe page. It is the same procedure for gray scale as well.  

[Sateesh Babu G]

@Ashwani Rai, Thanks for your response. I have followed the procedure described by dennis-chen in the post https://github.com/BVLC/caffe/issues/550

In create_imagenet.sh, I have passed --gray into convert_imageset as follow:

GLOG_logtostderr=1 $TOOLS/convert_imageset \

    --resize_height=$RESIZE_HEIGHT \

    --resize_width=$RESIZE_WIDTH \

    --gray \

    --shuffle \

    $TRAIN_DATA_ROOT \

    $DATA/Train.txt \

    $EXAMPLE/train_lmdb

In deploy.prototext, I have used input dimensions as follow:

name: "CaffeNet"

input: "data"

input_dim: 10

input_dim: 1

input_dim: 227

input_dim: 227

Finally in the python classifier wrapper, I am creating network as follow:

net = caffe.Classifier(MODEL_FILE, PRETRAINED, 

              mean = np.load('out.npy').mean(1).mean(1),

      raw_scale=255,

      image_dims=(256,256))

But still it is not working, all images are classified into one class.

[Sateesh Babu G]

Please can anyone explain me the procedure for gray scale images classification using Caffe. I have followed the below procedure but no success.

 I have followed the procedure described by dennis-chen in the post https://github.com/BVLC/caffe/issues/550

In create_imagenet.sh, I have passed --gray into convert_imageset as follow:

GLOG_logtostderr=1 $TOOLS/convert_imageset \

    --resize_height=$RESIZE_HEIGHT \

    --resize_width=$RESIZE_WIDTH \

    --gray \

    --shuffle \

    $TRAIN_DATA_ROOT \

    $DATA/Train.txt \

    $EXAMPLE/train_lmdb

In deploy.prototext, I have used input dimensions as follow:

name: "CaffeNet"

input: "data"

input_dim: 10

input_dim: 1

input_dim: 227

input_dim: 227

Finally in the python classifier wrapper, I am creating network as follow:

net = caffe.Classifier(MODEL_FILE, PRETRAINED, 

               mean = np.load('out.npy').mean(1).mean(1),

       raw_scale=255,

       image_dims=(256,256))

But still it is not working, all images are classified into one class.

Thanks in advance!!!!!!!