Torch: Model fast when learning/testing, slow when using it -


i have issue using learned model torch.

i followed howto http://code.cogbits.com/wiki/doku.php?id=tutorial_supervised train model. fine, model trained , have corrects results when use model. it's slow !

the testing part training this:

model:evaluate()  -- test on test data print('==> testing on test set:') t = 1,testdata:size()    -- disp progress    xlua.progress(t, testdata:size())     -- new sample    local input = testdata.data[t]    if opt.type == 'double' input = input:double()    elseif opt.type == 'cuda' input = input:cuda() end    local target = testdata.labels[t]     -- test sample    local pred = model:forward(input)    confusion:add(pred, target) end  -- timing time = sys.clock() - time time = time / testdata:size() print("\n==> time test 1 sample = " .. (time*1000) .. 'ms') 

i have following speed recorded during testing:

==> time test 1 sample = 12.419194088996ms

(of course vary, it's ~12ms).

i want use learned model on others images, did in simple , new script:

(... requires)  torch.setnumthreads(8) torch.setdefaulttensortype('torch.floattensor')  model = torch.load('results/model.net') model:evaluate()  (... image loading, resizing , normalization)  local time = sys.clock()  local result_info = model:forward(cropped_image:double())  print("==> time test 1 frame = " .. (sys.clock() - time) * 1000 .. "ms") 

the time spent bigger, have following output: ==> time test 1 frame = 212.7647127424ms

i tested more 1 image, resizing , normalization outside clock's measurements, , have > 200ms / image.

i don't understand i'm doing wrong , why code slower during training/testing.

thanks !


Comments

Popular posts from this blog

c++ - No viable overloaded operator for references a map -

java - Custom OutputStreamAppender not run: LOGBACK: No context given for <MYAPPENDER> -

java - Cannot secure connection using TLS -