test model get very low accuracy

[iodncookie]

hi,i had try my model use 90 epochs.here is my result:

but when i run test.py,i get very low accuracy,

i am so confused.i wonder know is something wrong?

[iodncookie]

@dontfollowmeimcrazy

[matteo-dunnhofer]

Hi @iodncookie, sorry for the late reply!

Are you sure to have loaded the right checkpoint of the model?

[iodncookie]

@dontfollowmeimcrazy
thanks for your reply。
yes, i am sure loaded the right checkpoint of the model,and here is my test.py code:

import os.path
import tensorflow as tf
import train_util as tu
from models import alexnet
import numpy as np
import logging
import os
os.environ['CUDA_VISIBLE_DEVICES'] = '1'

logging.basicConfig(level=logging.INFO,
filename='test_loger.log',
filemode='w',
format='%(asctime)s %(message)s ',
datefmt='%Y-%m-%d %H:%M:%S',
)

def test(
top_k,
k_patches,
display_step,
imagenet_path,
ckpt_path):

test_images = sorted(os.listdir(os.path.join(imagenet_path, 'val')))
test_labels = tu.read_test_labels(os.path.join(imagenet_path, 'data/ILSVRC2012_validation_ground_truth.txt'))

test_examples = len(test_images)

x = tf.placeholder(tf.float32, [None, 224, 224, 3])
y = tf.placeholder(tf.float32, [None, 1000])

# _, pred = alexnet.classifier(x, 1.0)
_, pred = alexnet.darknet53_classifier(x,False)
# calculate the average precision of the crops of the image
avg_prediction = tf.div(tf.reduce_sum(pred, 0), k_patches)

# accuracy
top1_correct = tf.equal(tf.argmax(avg_prediction, 0), tf.argmax(y, 1))
top1_accuracy = tf.reduce_mean(tf.cast(top1_correct, tf.float32))

topk_correct = tf.nn.in_top_k(tf.stack([avg_prediction]), tf.argmax(y, 1), k=top_k)
topk_accuracy = tf.reduce_mean(tf.cast(topk_correct, tf.float32))

saver = tf.train.Saver()

with tf.Session(config=tf.ConfigProto()) as sess:
	saver.restore(sess, os.path.join(ckpt_path, 'darknet_prune_77_0.7422.ckpt'))

	total_top1_accuracy = 0.
	total_topk_accuracy = 0.

	for i in range(test_examples):
		# taking a few patches from an image
		image_patches = tu.read_k_patches(os.path.join(imagenet_path, 'val', test_images[i]), k_patches)
		label = test_labels[i]

		top1_a, topk_a = sess.run([top1_accuracy, topk_accuracy], feed_dict={x: image_patches, y: [label]})
		total_top1_accuracy += top1_a
		total_topk_accuracy += topk_a

		if i % display_step == 0:
			info = 'Examples done: {:5d}/{} ---- Top-1: {:.4f} -- Top-{}: {:.4f}'.format(i + 1, test_examples, total_top1_accuracy / (i + 1), top_k, total_topk_accuracy / (i + 1))
			print(info)
			logging.info(info)
	print ('---- Final accuracy ----')
	info = 'Top-1: {:.4f} -- Top-{}: {:.4f}'.format(total_top1_accuracy / test_examples, top_k, total_topk_accuracy / test_examples)
	print(info)
	logging.info(info)
	info = 'Top-1 error rate: {:.4f} -- Top-{} error rate: {:.4f}'.format(1 - (total_top1_accuracy / test_examples), top_k, 1 - (total_topk_accuracy / test_examples))
	print(info)
	logging.info(info)

if name == 'main':
TOP_K = 5
K_PATCHES = 5
DISPLAY_STEP = 10
IMAGENET_PATH = '/home/yons/work/imagenet'
CKPT_PATH = 'ckpt-darknet-prune'

test( 
	TOP_K, 
	K_PATCHES, 
	DISPLAY_STEP,
	IMAGENET_PATH,
	CKPT_PATH)

[ghost]

I have the same problem

[matteo-dunnhofer]

Hi @iodncookie
Are you training and testing a different model than the AlexNet present in the repo? I noticed this line in the code you pasted
_, pred = alexnet.darknet53_classifier(x,False)

@codinggirl-abc are you doing something similar?

[iodncookie]

Hi @iodncookie
Are you training and testing a different model than the AlexNet present in the repo? I noticed this line in the code you pasted
_, pred = alexnet.darknet53_classifier(x,False)

@codinggirl-abc are you doing something similar?

yes,i am training a prune darknet model. i am still strugle with this problem.

[matteo-dunnhofer]

Have you tried to train AlexNet? Does it give you the same problem?

[iodncookie]

Have you tried to train AlexNet? Does it give you the same problem?
not yet.train a model almost take me two weeks,so i just directly train my own model.