Hi @Boyiliee ,
Can you please tell what backprapogation algo. you used for gradient calculation
I went through your paper but did not find any mention of backprapogation algo./function which you
used for training your AOD model although other training parameters such as learning rate, momentum, weight decay etc... are mentioned

Hi @Boyiliee
while training AOD-Net with NYU2 database "Loss is not converging"
I prepared train script but during training loss is not converging
I am training network with NYU2 database with 27, 256 as mentioned in your paper
I have used your test_template.prototxt as training.prototxt and I am training network with 150000 iteration ~ ((27,567/batchsize=8)*40 epoch)
as Euclidean loss layer is used in prototxt loss is not converging over the period it remain same around 15000 to 16000
upon training completion I even checked for inference with trained model_iter_150000.caffemodel instead of your AOD_Net.caffemodel but it is producing completely white image as output instead of giving dehazed image

below is my solver.prototxt content

net:"Training.prototxt"
base_lr: 0.001
lr_policy: "fixed"
display: 20
max_iter: 150000
momentum: 0.9
weight_decay: 0.0001
snapshot: 15000
snapshot_prefix: "models/model"
solver_mode: GPU
type: "SGD"

Training.prototxt is same as test_template.prototxt with below modification
input_dim: 1 -----> input_dim: {batchSize} for data/label layer
and for each conv layer i included weight_filler in convolution_param
weight_filler {{
type: "gaussian"
}}

My train script

import os
import numpy as np
from pylab import *
import re
import random
import cv2
print cv2.version
import ntpath
ntpath.basename('a/b/c')
def path_leaf(path):
head, tail = ntpath.split(path)
return tail or ntpath.basename(head)

Train_DIR = '../data/AODtrain/training/'
Label_DIR = '../data/AODtrain/original/'

Network Training parameters for input Image data

height = 480
width = 640
batch = 8 # batch size

import sys
sys.path.append("/home/ubuntu/Tools/caffe/python/")
import caffe

def EditFcnProto(templateFile, height, width, batch_size):
with open(templateFile, 'r') as ft:
template = ft.read()
# print templateFile
outFile = 'Training.prototxt'
with open(outFile, 'w') as fd:
fd.write(template.format(height=height,width=width,batchSize=batch_size))

def createBatch(img_dir, label_dir, batch_size):
batchdata = []
labelbatchdata = []
for i in range(batch_size):
fname = random.choice(os.listdir(img_dir))
imagepath = Train_DIR + fname
# print fname
# print imagepath
labelpath = label_dir + fname.split('')[0] + '' + fname.split('_')[1] + '.jpg'
# print labelpath

    npstore = caffe.io.load_image(imagepath)
    labelstore = caffe.io.load_image(labelpath)

    data = npstore
    data = data.transpose((2, 0, 1))
    batchdata.append(data)
    label = labelstore
    label = label.transpose((2, 0, 1))
    labelbatchdata.append(label)
return batchdata, labelbatchdata

def train():
caffe.set_mode_gpu()
caffe.set_device(0)

# training.prototxt is same as test_template.prototxt with only modification for data/label layer is
# input_dim: 1 -----> input_dim: {batchSize}
templateFile = 'train_template.prototxt'
EditFcnProto(templateFile, height, width, batch)

solver = caffe.SGDSolver('solver.prototxt')
# solver = caffe.get_solver('solver.prototxt')



niter = 150000
train_loss = np.zeros(niter)

f = open('loss.txt', 'w')

for it in range(niter):
    batchdata, labelbatchdata = createBatch(Train_DIR, Label_DIR, batch)
    solver.net.blobs['data'].data[...] = batchdata;
    solver.net.blobs['label'].data[...] = labelbatchdata;
    solver.step(1)
    train_loss[it] = solver.net.blobs['loss'].data
    f.write('{0: d} '.format(it))
    f.write('{0: f}\n'.format(train_loss[it]))

f.close()

if name == 'main':
train()

snapshot of output

I0530 15:41:02.473088 2152 sgd_solver.cpp:112] Iteration 51940, lr = 0.001
I0530 15:41:07.138154 2152 solver.cpp:239] Iteration 51960 (4.28732 iter/s, 4.66492s/20 iters), loss = 157563
I0530 15:41:07.138191 2152 solver.cpp:258] Train net output #0: loss = 157563 (* 1 = 157563 loss)
I0530 15:41:07.138197 2152 sgd_solver.cpp:112] Iteration 51960, lr = 0.001
I0530 15:41:11.762470 2152 solver.cpp:239] Iteration 51980 (4.32514 iter/s, 4.62412s/20 iters), loss = 156092

I am confused what is going wrong here loss is not converging, please tell me what I am doing wrong and suggest how to proceed
basically I am trying to reproduce your paper for learning

Thanks

I can't find the test dataset in the authors' website.Can you provide the test dataset and the code of PSNR and SSIM?Thank you very much!

I used AOD-Net to train and test the cityscapes_foggy dataset that including 500 training set, when I train 2epoch, the loss about 0.001. The results is not good.
Can you give me some advice to train the model in the dataset. thanks.
the dataset Link: http://people.ee.ethz.ch/~csakarid/SFSU_synthetic/

hello ：
I am very glad to see your public your model, I have already seen your paper, there is a question to ask you, you mentioned in the paper “It is also found helpful to clip the gradient to constrain the
norm within [−0.1, 0.1] ”.It means the solve parameter which called "clip_gradient" is among [-0.1,0.1]

What is the model size and no. of Parameters of the trained AOD model?

Hi,
Thanks for your code and paper! But I have several questions after looking at your code.

1. Why do u choose Relu layer as the last layer since Relu is not bounded for producing the image?
2. In your paper, the b is the constant bias with the default value 1, may I know why you choose b as 1? Combing with the first question, you add 1 and then apply Relu layer, am I correct?

Thanks in advance!

When I tried to run the file test.py, an error occured as folows:

Traceback (most recent call last):
  File "test/test.py", line 78, in <module>
    main();
  File "test/test.py", line 74, in main
    test()
  File "test/test.py", line 58, in test
    net.blobs['data'].data[...] = batchdata;
ValueError: could not broadcast input array from shape (3,540,360) into shape (3,600,400)

It seemed that the error was caused by the transpose operation in test
data = data.transpose((2, 0, 1))

and the dimension of the image was somehow not adaptive to the code.
But I changed the line 43 in test
npstore = caffe.io.load_image('data/img/%s.jpg'% imagename)
to
npstore = caffe.io.load_image('data/img/gugong.jpg')

The code ran normally. I want to know what led to the error mentioned above. Thank you!

from utils import logger, weight_init
from config import get_config
from data import HazeDataset

这几个代码好像没有提供 运行出错 请问是什么原因呢

@Boyiliee Thanks for sharing your work publicly I went through your paper and wanted to reproduce it ....I want to to train AOD-NET with my own data
I am new to caffe I can see test.py script but there is no train.py can you please help me on how to train AOD-NET, I will really appreciate your help