Skip to content

Instantly share code, notes, and snippets.

View davidbau's full-sized avatar

David Bau davidbau

944 followers · 0 following
View GitHub Profile
@davidbau
davidbau / +lightbox.html
Created January 14, 2023 21:07
lightbox.html, a page for automatically showing all images in a directory on an Apache server.
<!DOCTYPE html>
<html>
<!--
+lightbox.html, a page for automatically showing all images in a
directory on an Apache server. Just copy it into the directory.
Works by scraping the default directory HTML at "./" - David Bau.
-->
<head>
<script src="https://cdn.jsdelivr.net/npm/[email protected]/dist/vue.js"
integrity="sha256-CMMTrj5gGwOAXBeFi7kNokqowkzbeL8ydAJy39ewjkQ="
@davidbau
davidbau / runningstats.py
Last active June 10, 2024 21:43
Running stats objects for pytorch: mean, variance, covariance, second-moment, quantiles, topk, and combinations.
'''
To use a runningstats object,
1. Create the the desired stat object, e.g., `m = Mean()`
2. Feed it batches via the add method, e.g., `m.add(batch)`
3. Repeat step 2 any number of times.
4. Read out the statistic of interest, e.g., `m.mean()`
Built-in runningstats objects include:
@davidbau
davidbau / nethook.py
Last active April 1, 2022 19:38
Utilities for instrumenting a pytorch model: Trace, TraceDict, subsequence, get_module, replace_module, get_parameter, set_requires_grad.
"""
Utilities for instrumenting a torch model. (David Bau)
Trace will hook one layer at a time.
TraceDict will hook multiple layers at once.
subsequence slices intervals from Sequential modules.
get_module, replace_module, get_parameter resolve dotted names.
set_requires_grad recursively sets requires_grad in module parameters.
"""
@davidbau
davidbau / simpleganpaint.ipynb
Last active July 14, 2022 19:34
SimpleGanPaint.ipynb
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
@davidbau
davidbau / explorefeaturebyclick.ipynb
Created October 20, 2019 22:14
ExploreFeatureByClick.ipynb
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
@davidbau
davidbau / seeing_images.ipynb
Created October 17, 2019 08:03
seeing_images.ipynb
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
@davidbau
davidbau / fooling.html
Created February 14, 2019 06:07
an example vue HTML file that displays images and metadata
<!DOCTYPE html>
<html>
<!-- an example vue HTML file that displays images and metadata -->
<head>
<script src="https://cdn.jsdelivr.net/npm/[email protected]/dist/vue.js"
integrity="sha256-CMMTrj5gGwOAXBeFi7kNokqowkzbeL8ydAJy39ewjkQ="
crossorigin="anonymous"></script>
<script src="https://cdn.jsdelivr.net/npm/[email protected]/lodash.js"
integrity="sha256-qwbDmNVLiCqkqRBpF46q5bjYH11j5cd+K+Y6D3/ja28="
crossorigin="anonymous"></script>
@davidbau
davidbau / featurefolder.py
Last active January 8, 2019 05:50
import os
import torch.utils.data as data
from torchvision.datasets.folder import default_loader, is_image_file
from PIL import Image
def grayscale_loader(path):
with open(path, 'rb') as f:
return Image.open(f).convert('L')
class FeatureFolder(data.Dataset):
@davidbau
davidbau / autoimport.py
Created August 26, 2018 08:50
Used to eval python expressions containing arbitrary fully-qualified package names, automatically importing packages as needed.
from collections import defaultdict
from importlib import import_module
def autoimport_eval(term):
'''
Used to eval any expression containing fully-qualified names, such as
'torchvision.models.alexnet(pretrained=True)' with automatic import of
global module names.
'''
@davidbau
davidbau / download_mp.py
Last active February 21, 2018 15:19
Script for downloading and formatting miniplaces in pytorch ImageFolder format
#!/usr/bin/env python2.7
# Script to create simple flat pytorch ImageFolder folder hierarchy
# of training and validation images for miniplaces. Each category
# name is just a folder name (numbered in alphabetical order as in
# the original miniplaces), and both train and val images are places
# directly inside a single level of folders with the flat cateogry names.
import shutil, os, tarfile