Low accuracy using my own implemented resnet on cifar100

N_Hunter · June 11, 2018, 2:24pm

Hi,everyone, Im kind new to pytorch, or any deep learning frameworks. I wrote my own residual network, and trained on cifar100 dataset, but the accuracy was around 1%, basically like random guess.
here is my github repo:

here is my resnet.py file:

github.com

weiaicunzai/pytorch-cifar-100/blob/master/models/resnet.py

"""resnet in pytorch



[1] Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun 
    Deep Residual Learning for Image Recognition
    https://arxiv.org/abs/1512.03385
"""

import torch
import torch.nn as nn


class BasicBlock(nn.Module):
    """Basic Block for resnet 18 and resnet 34

    """

    #BasicBlock and BottleNeck block 
    #have different output size

This file has been truncated. show original

here is my train.py file:

github.com

weiaicunzai/pytorch-cifar-100/blob/master/train.py

# train.py
#!/usr/bin/env	python3

""" train network using pytorch
"""

from torch.utils.data import DataLoader
import torch

from dataset import CIFAR100Train 
from models.resnet import *
import torch.optim as optim
from torch.autograd import Variable

from settings import *

cifar100_train = DataLoader(CIFAR100Train(g_cifar_100_path), shuffle=True, num_workers=2, batch_size=16)

net = ResNet101()

This file has been truncated. show original

Im using python3.5 and pytroch 0,31

so could you please tell me where the bug is, or how can I find the bug in a neuron network in general, or
just some hints, anything would be helpful.

thanks in advance, have a nice day.

N_Hunter · June 12, 2018, 4:29am

never mind, only trained for one epoch caused the problem.