Introduction

“On Instagram” reimagines Bill Cunningham’s “On the Street” column for the digital age, showcasing fashion trends from Instagram, curated and analyzed using advanced deep learning technology.

“On the Street” by Bill Cunningham

Bill Cunningham, a renowned American fashion photographer, immortalized the vibrant streets of New York City in his iconic New York Times column On the Street. Known for his democratic approach to fashion photography, Bill captured the evolving urban tapestry and individual style expressions on his bicycle.

Image source: It Was Bill Cunningham’s New York; We Were Just Living In It

His work, more than just images, narrated the fashion zeitgeist of everyday New Yorkers. From Fifth Avenue’s high fashion to downtown’s eclectic mixes, he celebrated the diversity of street style. “On the Street” not only transcended fashion journalism norms but also became a visual diary reflecting the city’s pulse through fashion. Cunningham’s enduring legacy continues in the column, inspiring readers with the dynamic, ever-changing world of street fashion.

Image source: On the Street 2016-05-01

ALL ASWIRL Pleated skirts started showing up last autumn but were quickly hidden under winter coats. They are back now, in multiple colors, fabrics and lengths. They are not a major trend; more like a caprice of fashion that women clearly love wearing. So does the artist Timothy John, bottom row, who creates his own sophisticated combinations.

Image source: On the Street 2016-05-15

RIPS AND TEARS The concept of radical chic has been evolving for a couple of years and is now heading into formal evening wear. Guests at black-tie affairs have appeared wearing their decon structed jeans — most glaringly at the Met Gala, planned by Anna Wintour, who suggests that men wear white tie. One guest wore a bejeweled jacket and his knees exposed in torn jeans, left. Is this where evening wear is headed? Long skirts are also being slashed, left. The label Hood by Air is showing skirts and pants cut at the knee, the bottom half suspended by ribbons, top. Is this a reflection of the tearing apart of established political parties?

From “On the Street” to “On Instagram”

My introduction to Bill Cunningham was through the documentary Bill Cunningham New York. I was profoundly impressed by Bill’s remarkable ability to recognize fashion patterns. His knack for uncovering, curating, and summarizing emerging trends was like finding a hidden gem in a chaotic world.

Image source: Bill Cunningham New York (2010)

When I was starting out in machine learning in 2016, I noticed similarities between Bill’s approach and clustering algorithms. Even though tasks like visual embedding, which are straightforward today, were challenging back then, I was eager to explore the possibilities of what could be accomplished with machine learning.

However, gathering street fashion photographs was beyond my reach. As someone immersed in the digital age, I naturally gravitated towards Instagram for insights into fashion.

Recap of My Group Project in 2016

With my team’s support, this ambitious concept evolved into an important group project in my master’s program. We aimed to semi-automate Bill Cunningham’s technique of discerning fashion trends by clustering Instagram posts from notable fashion influencers.

We compiled a modest dataset from 90 fashion influencers on Instagram. Due to our limited understanding of deep learning and the absence of readily available open-source deep neural networks, we opted for RGB-SIFT for feature extraction. To minimize background distractions, we first used a human body detection algorithm to isolate the human figure in the images, assuming they were present in most photos. Utilizing these features, we applied basic KMeans clustering to identify significant trends.

To present our results with visual appeal comparable to Bill’s posters, we straightforwardly replicated his layout. Crafting names for the posters in his style proved difficult due to my limited writing and fashion expertise. In contrast, today, it’s clear that seeking ChatGPT’s help for such creative tasks would be beneficial.

Reflecting on our project, our approach and the dataset might appear basic, yet the outcomes were surprisingly perceptive. We successfully identified two clusters based on color and three based on style for the months of January through March.

Revisiting the Project in 2023

Over the past seven years, the deep learning field has experienced a significant evolution. The introduction of open-source foundation models for visual embedding and open-set object detection, along with a host of other advanced techniques, has substantially advanced the field. Simultaneously, my skills in machine learning and computer vision have grown.

Despite these changes, my interest in this project remains steadfast. With the aid of ChatGPT, bolstered by a suite of potent tools and capabilities, I am excited to independently revisit this project. By employing modern techniques, I am keen to uncover the potential achievements that these years of technological progress have made possible.