Is there any way I can use pretrained Imagenet models in Fast.ai to label images?

xjdeng · July 27, 2018, 4:46am

Basically do the following:

Take an image as an input.

Output a list of possible classes and their probabilities as output:

Without doing any training (basically using the pretrained Imagenet models)

Example:

Tree: 0.3
Person: 0.3
Wolf: 0.3
Mushroom: 0.05
Frog: 0.05

I know how to do that in Keras, example here.

Is there any way to achieve THE EXACT SAME RESULTS as above using Pytorch or Fast.ai?

The closest thing I’ve found to this in Pytorch.

xjdeng · April 11, 2019, 2:00pm

Yes, use the classify() function here:

github.com

xjdeng/Easy_Image/blob/master/Easy_Image/classify_pytorch.py

# Based loosely off of 
# http://blog.outcome.io/pytorch-quick-start-classifying-an-image/
# https://discuss.pytorch.org/t/trouble-getting-pretrained-resnet152-to-classify-images-properly/20073

import os, json
from torchvision import models, transforms
from torch.autograd import Variable
from PIL import Image
import cv2
import numpy as np

normalize = transforms.Normalize(
   mean=[0.485, 0.456, 0.406],
   std=[0.229, 0.224, 0.225]
)
preprocess = transforms.Compose([
   transforms.Scale(256),
   transforms.CenterCrop(224),
   transforms.ToTensor(),
   normalize

This file has been truncated. show original