These are chat archives for beniz/deepdetect

15th
Nov 2017
jubeenshah
@jubeenshah
Nov 15 2017 12:44 UTC
Hello everyone I'm new to Deepdetect and a novice at it. I've followed the tutorial and set up the image classifier at title

however when I'm testing the classifier I'm getting an error

{"status":{"code":400,"msg":"BadRequest","dd_code":1006,"dd_msg":"Service Bad Request Error"}}

curl http://localhost:8020/info

{"status":{"code":200,"msg":"OK"},"head":{"method":"/info","version":"0.1","branch":"master","commit":"978401f3d1f23a327d0ebfef24cb0a0d7c543c6e","services":[{"mllib":"caffe","description":"image classification service","name":"imageserv"}]}}

could someone please help, I don't know whether this chatroom can be used for queries.
Emmanuel Benazera
@beniz
Nov 15 2017 13:59 UTC
Hi, post your full testing call
jubeenshah
@jubeenshah
Nov 15 2017 14:00 UTC
Hello, I guess I hadn't started the service, I'm currently trying to start the service, will let you know how it goes.. Thanks a lot for the quick response though
Okay so I downloaded the ResNet-50-model.caffemodel and placed it in the model directory. The next command that I'm running is .dede, however I'm not getting any response 1) how do I know that the deep detect server is running?
2)curl -X PUT "http://localhost:8020/services/imageserv" -d '{ "mllib":"caffe", "description":"image classification service", "type":"supervised", "parameters":{ "input":{ "connector":"image" }, "mllib":{ "template”:”resnet_50”, "nclasses":1000 } }, "model":{ "templates":"/home/shreos/Documents/deepdetect/templates/caffe/", "repository":"/home/shreos/Documents/deepdetect/models/imgnet/" } }'
jubeenshah
@jubeenshah
Nov 15 2017 14:05 UTC
I'm running this code however
rperdon
@rperdon
Nov 15 2017 14:06 UTC
Once running, it should provide a prompt of the status
jubeenshah
@jubeenshah
Nov 15 2017 14:06 UTC
I'm getting -- this error {"status":{"code":400,"msg":"BadRequest"}}
no @rperdon it doesn't
rperdon
@rperdon
Nov 15 2017 14:09 UTC
INFO - 14:10:15 - Running DeepDetect HTTP server on localhost:8080
If dede is running correctly, you should get that initial prompt
jubeenshah
@jubeenshah
Nov 15 2017 14:09 UTC
oh.. okay I'll check that out again. thanks
rperdon
@rperdon
Nov 15 2017 14:09 UTC
that terminal window is now your status window
you need to run your curl commands from a second terminal window
I'm running it from his docker image
jubeenshah
@jubeenshah
Nov 15 2017 14:10 UTC
Okay! I was also not following that.... thanks :smile:
rperdon
@rperdon
Nov 15 2017 14:35 UTC
Just a quick question on the training images, you mentioned gif is not supported (due to animated gifs I think). Does DD support all other image types?
Emmanuel Benazera
@beniz
Nov 15 2017 14:45 UTC
all other types supported by opencv
jubeenshah
@jubeenshah
Nov 15 2017 14:48 UTC

I guess I kind of figured out the error please have a look at this CMake Error at CMakeLists.txt:110 (find_package):
By not providing "FindOpenCV.cmake" in CMAKE_MODULE_PATH this project has
asked CMake to find a package configuration file provided by "OpenCV", but
CMake did not find one.

Could not find a package configuration file provided by "OpenCV" with any
of the following names:

OpenCVConfig.cmake
opencv-config.cmake

Add the installation prefix of "OpenCV" to CMAKE_PREFIX_PATH or set
"OpenCV_DIR" to a directory containing one of the above files. If "OpenCV"
provides a separate development package or SDK, be sure it has been
installed.

-- Configuring incomplete, errors occurred!

rperdon
@rperdon
Nov 15 2017 15:10 UTC
While I am using deepdetect to train right now, I'm still working on discovering how to get some good accuracy data consistent to our previous model work. I would like to still help on discovering the differences in how we can get dd to load images the same as other models like the yahoo_nsfw, models from digits, and other caffe models without worries that there would be divergences in the classification results. I'm wondering if you have more ideas for debugging I can test out.
rperdon
@rperdon
Nov 15 2017 18:42 UTC
Any tricks to getting a convnet to train?
{"status":{"code":500,"msg":"InternalError","dd_code":1007,"dd_msg":"CHECK failed: (index) < (size()): "}}
Emmanuel Benazera
@beniz
Nov 15 2017 18:53 UTC
Not sure I know... Maybe wrong number of classes...
rperdon
@rperdon
Nov 15 2017 18:54 UTC
weird, still same data set. I was reading on convnet's strengths in binary classification and wanted to try it out
I'm trying to base it off the settings provided in the examples page you have
rperdon
@rperdon
Nov 15 2017 19:08 UTC
{"status":{"code":500,"msg":"InternalError","dd_code":1007,"dd_msg":"src/caffe/layers/memory_datalayer.cpp:116 / Check failed (custom): data"}}
Emmanuel Benazera
@beniz
Nov 15 2017 19:09 UTC
Wrong .lmdb file or something related to your test or deploy inputs...
rperdon
@rperdon
Nov 15 2017 19:11 UTC
 curl -X POST "http://localhost:9999/train" -d '{
   "service":"ddanimemodel",
   "async":false,
   "parameters":{
     "mllib":{
       "gpu":true,
       "net":{
         "batch_size":32
       },
       "solver":{
         "test_interval":500,
         "iterations":10000,
         "base_lr":0.001,
         "solver_type":"SGD"
       }
     },
     "input":{
       "connector":
       "image",
       "test_split":0.1,
       "shuffle":true
     },
     "output":{
       "measure":["mcll","f1"]
     }
   },
   "data":["/source"]
 }'
curl -X PUT "http://localhost:9999/services/ddanimemodel" -d '{
"mllib":"caffe",
"description":"anime classifier",
"type":"supervised",
"parameters":{
"input":{
"connector":"image",
"width":64,
"height":64
},
"mllib":{
"template":"convnet",
"nclasses":2,
"layers":["1CR32","1CR64","1CR128","1024"],
"dropout":0.2
}
},
"model":{
"templates":"../templates/caffe/",
"repository":"/source"
}
}'
Emmanuel Benazera
@beniz
Nov 15 2017 19:12 UTC
Use templates instead of custom convnets
They won't work better
Use resnet_50
rperdon
@rperdon
Nov 15 2017 19:13 UTC
I tried the resnet_50; I still had 50% accuracy results
Emmanuel Benazera
@beniz
Nov 15 2017 19:13 UTC
,
rperdon
@rperdon
Nov 15 2017 19:13 UTC
0.001 lr and 0.1, and 0.0001 tested