Code repository for "It's About Time: Analog clock Reading in the Wild"

Last update: Nov 10, 2022

Related tags

Text Data & NLP itsabouttime

Overview

it's about time

Code repository for "It's About Time: Analog clock Reading in the Wild"

Packages required: pytorch (used 1.9, any reasonable version should work), kornia (for homography), einops, scikit-learn (for RANSAC), tensorboardX (for logging)

Using pretrained model:

prediction python predict.py will predict on your data (or by default, whatever is in data/demo). This does assume the images being already cropped, we use CBNetv2. (you could instead add something like a yolov5 to the code if you prefer not installing anything extra).
evaluation python eval.py (requires dataset) should return the numbers reported in the paper

Training:

sh full_cycle.sh should do the job
if you want to do it individually, then do use
- train.py train on SynClock
- generate_pseudo_labels.py use the model to generate pseudo labels for timelapse
- train_refine.py train on SynClock+timelapse.
- The latter two can be repeated iteratively.

Dataset (Train):

SynClock is generated on the fly (via SynClock.py)
Timelapse will be uploaded later.

Dataset (Eval):

COCO and OpenImages: The .csv files in data/ contains the image ids, predicted bbox's (by CBNetV2), gt bbox's, and the manual time label. We will upload this subset later for convenience, but if you already have the respective datasets it should already work.
Clock Movies do not contain bbox's. We may not be able to release the data directly due to copyright, but the csv files do contain the image file names, and they are scraped from https://theclock.fandom.com/wiki/Special:NewFiles

Note: src/cyclic_ransac.py is adapted from the source code of scikit-learn (authored by Johannes Schönberger under BSD 3 clause license), to fit a sawtooth wave for cyclic linear data.

Coming soon (early December):

video
dataset
detection

Comments

On the problem of time-lapse photographic data set

Thank you for your excellent work. I'm a beginner who just came into contact with this direction. I encountered a problem in the program of labeling delayed video. I didn't find the relevant code for video processing, and when running the source code, I found that the directory of the published timelapse dataset didn't correspond to the given timelapse.txt label file. I thought it was the problem of my directory setting. Please answer it

opened by zw-2503687 0
Some files which names can be find in 'lable.txt' cannot find on Timelapse datasets
Hi! Thanks for your work and it's meaningful and amazing. However, I meet some problems when I run the code you provided.

About generate_pseudo_labels.py, the input needs to be images not video provided on timelapse dataset. So do u do other actions ?

Timelapse data cannot corresponde with released label file (timelapse_label.txt)

During training process, I do not understand the meaning to generate pseudo-labels for training dataset, would't it be better to calibrate it with a well-trained models?
opened by Vincen-31 1
Some question about train STN

Hi, thank you for your amazing work. I am inspired by your work about alignment the clock. I have tried to use STN for alignment my pointer meter as followings. I picked front meter of each category as a standard meter. And online random homography matrix was generated by code to apply on these standard meters for training STN. My batchsize is 32, and iteration about 120000, trian data is 10000, but the results is not good, and the loss is about 0.5 finally. Could you give my some advice about this task?

Thanks very much.

opened by xywlpo 4
About train_refine

Thanks for such a great job! I am implementing train_fine and have doubts about the following code: img = torch.cat([img, img2], 0) hour = torch.cat([hour, hour2], 0) minute = torch.cat([minute, minute2], 0) Why overlay data

opened by Totorol 1
training time

Dear authors,

Thanks for this wonderful work! Could you please let me know your training time (e.g. at how many GPUS) and the total epoch number? Did you set an early stop scheme?

opened by catherine-qian 1
"kornia.geometry.transform.warp_perspective"

Thanks for sharing the excellent job! In achieving STN, instead of the pytorch tutorials of STN, the codes adopt "kornia.geometry.transform.warp_perspective" to generate new images. I wonder why kornia is adopted, and the differences between the methods.

opened by cuishuhao 2

Code repository for "It's About Time: Analog clock Reading in the Wild"

Related tags

Overview

it's about time

Comments

On the problem of time-lapse photographic data set

Some files which names can be find in 'lable.txt' cannot find on Timelapse datasets

Some question about train STN

About train_refine

training time

"kornia.geometry.transform.warp_perspective"

Owner

This repository will contain the code for the CVPR 2021 paper "GIRAFFE: Representing Scenes as Compositional Generative Neural Feature Fields"

This repository contains all the source code that is needed for the project : An Efficient Pipeline For Bloom’s Taxonomy Using Natural Language Processing and Deep Learning

This repository contains the code, data, and models of the paper titled "XL-Sum: Large-Scale Multilingual Abstractive Summarization for 44 Languages" published in Findings of the Association for Computational Linguistics: ACL 2021.

This repository contains the code for EMNLP-2021 paper "Word-Level Coreference Resolution"

This repository contains the official release of the model "BanglaBERT" and associated downstream finetuning code and datasets introduced in the paper titled "BanglaBERT: Combating Embedding Barrier in Multilingual Models for Low-Resource Language Understanding".

Officile code repository for "A Game-Theoretic Perspective on Risk-Sensitive Reinforcement Learning"

Code repository for "It's About Time: Analog clock Reading in the Wild"

Repository of the Code to Chatbots, developed in Python

This project is part of Eleuther AI's quest to create a massive repository of high quality text data for training language models.

This repository describes our reproducible framework for assessing self-supervised representation learning from speech

NL-Augmenter 🦎 → 🐍 A Collaborative Repository of Natural Language Transformations

Implementing SimCSE(paper, official repository) using TensorFlow 2 and KR-BERT.

This repository contains data used in the NAACL 2021 Paper - Proteno: Text Normalization with Limited Data for Fast Deployment in Text to Speech Systems

This repository is home to the Optimus data transformation plugins for various data processing needs.

This is the main repository of open-sourced speech technology by Huawei Noah's Ark Lab.

Repository for Graph2Pix: A Graph-Based Image to Image Translation Framework

Plugin repository for Macast

The repository for the paper: Multilingual Translation via Grafting Pre-trained Language Models