why is the accuracy=0%? (Lenet Model)

[Chias JaJa]

I use the Lenet model and picture's height is 13, width is 79.

Total:1000000 pic , two classes.

Training Data is 800000 pic, Testing Data is 200000 pic, have two classes.

The accuracy could at least have 50%.

why is accuracy=0%?

solver.prototxt：

net: "examples/mytry/lenet_train_test.prototxt"

test_iter: 2000

test_interval: 5000

base_lr: 0.001

momentum: 0.9

weight_decay: 0.005

lr_policy: "inv"

gamma: 0.1

power: 0.75

display: 1000

max_iter: 50000

snapshot: 50000

snapshot_prefix: "examples/mytry/lenet"

solver_mode: CPU

train_test.prototxt：(The part of change)

data_param {

source: "examples/mytry/img_train_lmdb"

batch_size: 64

backend: LMDB

}

data_param {

source: "examples/mytry/img_test_lmdb"

batch_size: 100

backend: LMDB

}

inner_product_param {

num_output: 2 #change 2 instead of 10

weight_filler {

type: "xavier"

}

bias_filler {

type: "constant"

}

}

Result：

I0515 15:19:19.808574 3082 solver.cpp:340] Iteration 40000, Testing net (#0)

I0515 15:21:53.017040 3082 solver.cpp:408] Test net output #0: accuracy = 0

I0515 15:21:53.017160 3082 solver.cpp:408] Test net output #1: loss = 1.18128 (* 1 = 1.18128 loss)

I0515 15:21:53.190886 3082 solver.cpp:236] Iteration 40000, loss = 0.690025

I0515 15:21:53.190979 3082 solver.cpp:252] Train net output #0: loss = 0.690025 (* 1 = 0.690025 loss)

I0515 15:21:53.190997 3082 sgd_solver.cpp:106] Iteration 40000, lr = 1.9878e-06

I0515 15:23:57.154177 3082 solver.cpp:236] Iteration 41000, loss = 2.02741

I0515 15:23:57.154613 3082 solver.cpp:252] Train net output #0: loss = 2.02741 (* 1 = 2.02741 loss)

I0515 15:23:57.154708 3082 sgd_solver.cpp:106] Iteration 41000, lr = 1.95134e-06

I0515 15:26:02.495137 3082 solver.cpp:236] Iteration 42000, loss = 0.667163

I0515 15:26:02.495523 3082 solver.cpp:252] Train net output #0: loss = 0.667163 (* 1 = 0.667163 loss)

I0515 15:26:02.495620 3082 sgd_solver.cpp:106] Iteration 42000, lr = 1.9164e-06

I0515 15:28:08.543061 3082 solver.cpp:236] Iteration 43000, loss = 2.0478

I0515 15:28:08.543457 3082 solver.cpp:252] Train net output #0: loss = 2.0478 (* 1 = 2.0478 loss)

I0515 15:28:08.543597 3082 sgd_solver.cpp:106] Iteration 43000, lr = 1.88288e-06

I0515 15:30:15.492943 3082 solver.cpp:236] Iteration 44000, loss = 0.709838

I0515 15:30:15.493360 3082 solver.cpp:252] Train net output #0: loss = 0.709839 (* 1 = 0.709839 loss)

I0515 15:30:15.493500 3082 sgd_solver.cpp:106] Iteration 44000, lr = 1.8507e-06

I0515 15:32:20.894922 3082 solver.cpp:340] Iteration 45000, Testing net (#0)

I0515 15:34:54.668030 3082 solver.cpp:408] Test net output #0: accuracy = 0

I0515 15:34:54.668329 3082 solver.cpp:408] Test net output #1: loss = 1.17161 (* 1 = 1.17161 loss)

I0515 15:34:54.808615 3082 solver.cpp:236] Iteration 45000, loss = 0.691562

I0515 15:34:54.808704 3082 solver.cpp:252] Train net output #0: loss = 0.691562 (* 1 = 0.691562 loss)

I0515 15:34:54.808724 3082 sgd_solver.cpp:106] Iteration 45000, lr = 1.81978e-06

I0515 15:36:58.802649 3082 solver.cpp:236] Iteration 46000, loss = 0.668188

I0515 15:36:58.803174 3082 solver.cpp:252] Train net output #0: loss = 0.668188 (* 1 = 0.668188 loss)

I0515 15:36:58.803314 3082 sgd_solver.cpp:106] Iteration 46000, lr = 1.79003e-06

I0515 15:39:03.794693 3082 solver.cpp:236] Iteration 47000, loss = 0.71103

I0515 15:39:03.795116 3082 solver.cpp:252] Train net output #0: loss = 0.711029 (* 1 = 0.711029 loss)

I0515 15:39:03.795261 3082 sgd_solver.cpp:106] Iteration 47000, lr = 1.7614e-06

I0515 15:41:08.893612 3082 solver.cpp:236] Iteration 48000, loss = 0.663108

I0515 15:41:08.893971 3082 solver.cpp:252] Train net output #0: loss = 0.663108 (* 1 = 0.663108 loss)

I0515 15:41:08.894016 3082 sgd_solver.cpp:106] Iteration 48000, lr = 1.73381e-06

I0515 15:43:13.983155 3082 solver.cpp:236] Iteration 49000, loss = 0.671532

I0515 15:43:13.983533 3082 solver.cpp:252] Train net output #0: loss = 0.671532 (* 1 = 0.671532 loss)

I0515 15:43:13.983672 3082 sgd_solver.cpp:106] Iteration 49000, lr = 1.70721e-06

I0515 15:45:18.354498 3082 solver.cpp:461] Snapshotting to binary proto file examples/mytry/lenet_iter_50000.caffemodel

I0515 15:45:18.385054 3082 sgd_solver.cpp:269] Snapshotting solver state to binary proto file examples/mytry/lenet_iter_50000.solverstate

I0515 15:45:18.457027 3082 solver.cpp:320] Iteration 50000, loss = 2.02316

I0515 15:45:18.457109 3082 solver.cpp:340] Iteration 50000, Testing net (#0)

I0515 15:47:51.883457 3082 solver.cpp:408] Test net output #0: accuracy = 0

I0515 15:47:51.883920 3082 solver.cpp:408] Test net output #1: loss = 1.18242 (* 1 = 1.18242 loss)

I0515 15:47:51.884052 3082 solver.cpp:325] Optimization Done.

I0515 15:47:51.884162 3082 caffe.cpp:215] Optimization Done.