Closed mpkuse closed 6 years ago
Hi Manohar,
For general documentation, I would go to TrainAndTest/main.py and run it with the args: ' test -h'. Basically, you can create precision recall curves with your dataset if you place the query and database images in directories called live and memory, respectively. Then analyze the sequential overlap for your dataset (i.e., multiple database matches i-n/2, ..., i, ..., i+n/2 for query image i in order to choose the parameter n. Finally, choose the right arguments to choose the aforementioned parameters based on the help command as mentioned, and point to the model you wish to test.
On Mon, Jul 2, 2018 at 23:02 Manohar Kuse notifications@github.com wrote:
Hi, Can you suggest me how can I test your algorithm with my own sequence? So my sequence is a series of images very similar to gradens-point dataset. I wish to extract the scene descriptors from your method for each of the images and compare the loop detections.
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/rpng/calc/issues/4, or mute the thread https://github.com/notifications/unsubscribe-auth/AXHa_2hflJPGdMctQz-x1urBAiHVaHJnks5uCt61gaJpZM4VAOMf .
-- Sincerely, Nate Merrill Researcher -- Robot Perception and Navigation Group nmerrill@udel.edu https://www.linkedin.com/in/nate-merrill
So, I downloaded your caffe models in DeepLCD/calc_model
. Also untared your CampusLoopDataset.
Finally running,
python main.py test -m ../DeepLCD/calc_model/ -d test_data/CampusLoopDataset/
Traceback (most recent call last):
File "main.py", line 157, in <module>
plot(net_model_path=args.model_path, data_path=args.data_path, num_include=args.num_include, title=args.title, resize_net=args.rz_net, alexnet_proto_path=args.alex_proto, alexnet_weights=args.alex_weights)
File "/home/mpkuse/Downloads/calc/TrainAndTest/testNet.py", line 326, in plot
precisions, recalls, precision_alex, recall_alex, precision_hog, recall_hog, precision_dbow, recall_dbow = get_prec_recall(net_def_path, net_model_path, data_path, num_include, title, resize_net, alexnet_proto_path, alexnet_weights)
File "/home/mpkuse/Downloads/calc/TrainAndTest/testNet.py", line 131, in get_prec_recall
nets.append(caffe.Net(net_def_path,1,weights=m_path))
RuntimeError: Could not open file proto/deploy.prototxt
I can see it cannot find the prototxt file, Am I specifying the arguments wrongly? What is the correct way?
Right,
The code expects the CALC prototxt model definition file to be at that location listed in your error. A quick fix is to copy it there, but I need to add an arg for that.
On Mon, Jul 2, 2018 at 23:42 Manohar Kuse notifications@github.com wrote:
So, I downloaded your caffe models in DeepLCD/calc_model. Also untared your CampusLoopDataset.
Finally running,
python main.py test -m ../DeepLCD/calc_model/ -d test_data/CampusLoopDataset/
Traceback (most recent call last): File "main.py", line 157, in
plot(net_model_path=args.model_path, data_path=args.data_path, num_include=args.num_include, title=args.title, resize_net=args.rz_net, alexnet_proto_path=args.alex_proto, alexnet_weights=args.alex_weights) File "/home/mpkuse/Downloads/calc/TrainAndTest/testNet.py", line 326, in plot precisions, recalls, precision_alex, recall_alex, precision_hog, recall_hog, precision_dbow, recall_dbow = get_prec_recall(net_def_path, net_model_path, data_path, num_include, title, resize_net, alexnet_proto_path, alexnet_weights) File "/home/mpkuse/Downloads/calc/TrainAndTest/testNet.py", line 131, in get_prec_recall nets.append(caffe.Net(net_def_path,1,weights=m_path)) RuntimeError: Could not open file proto/deploy.prototxt I can see it cannot find the prototxt file, Am I specifying the arguments wrongly? What is the correct way?
— You are receiving this because you commented.
Reply to this email directly, view it on GitHub https://github.com/rpng/calc/issues/4#issuecomment-402004166, or mute the thread https://github.com/notifications/unsubscribe-auth/AXHa_7huvplrP0bad8qAMuC0RnELRRo0ks5uCug9gaJpZM4VAOMf .
-- Sincerely, Nate Merrill Researcher -- Robot Perception and Navigation Group nmerrill@udel.edu https://www.linkedin.com/in/nate-merrill
Ok! Thanks got it working !
I am trying to get the descriptors for my images. I am trying to do something similar to this: https://github.com/rpng/calc/blob/166f90c4e77637fc030911bc4f65a79bcaa2b0b8/TrainAndTest/testNet.py#L446 ie. your function view_forward_pass()
.
Can you confirm that net.blobs['descriptor'].data
gets me the descriptor for my image using your proposed auto-encoder model?
Can you also provide me the caffemodel you used for Alexnet (alexnet-proto and alexnet-model)? Was this model used? https://github.com/BVLC/caffe/tree/master/models/bvlc_alexnet
Can you specify exactly what you need to do? I have that function for viewing the data as a heat map as in our paper. I also have the entire C++ library in DeepLCD for performing image matching. If you want to know how to do it for your own python code then the function in testNet is a good example. The descriptor layer is in fact the layer for the descriptor.
On Wed, Jul 4, 2018 at 02:46 Manohar Kuse notifications@github.com wrote:
I am trying to get the descriptors for my images. I am trying to do something similar to this: https://github.com/rpng/calc/blob/166f90c4e77637fc030911bc4f65a79bcaa2b0b8/TrainAndTest/testNet.py#L446 ie. your function view_forward_pass().
Can you confirm that net.blobs['descriptor'].data gets me the descriptor for my image using your proposed auto-encoder model?
— You are receiving this because you commented.
Reply to this email directly, view it on GitHub https://github.com/rpng/calc/issues/4#issuecomment-402380248, or mute the thread https://github.com/notifications/unsubscribe-auth/AXHa_-DqPZAj3AJMbG0gv4z3skYB3f_Eks5uDGTFgaJpZM4VAOMf .
-- Sincerely, Nate Merrill Researcher -- Robot Perception and Navigation Group nmerrill@udel.edu https://www.linkedin.com/in/nate-merrill
I forgot to answer the last part of your question sorry. Yes that model from BVLC is the model we used for AlexNet in the paper.
Ok. Thanks for all your replies. I got your code working as I desired.
Hi, Can you suggest me how can I test your algorithm with my own sequence? So my sequence is a series of images very similar to gradens-point dataset. I wish to extract the scene descriptors from your method for each of the images and compare the loop detections.