Adobe Sneaks: Changing Outfits in eCommerce with AI

By Eric Mati­soff, Ana­lyt­ics & Data Sci­ence Evan­ge­list at Adobe

Once a year, a call goes out to thou­sands of employ­ees across Adobe’s offices around the world. Any­one in the com­pa­ny, from engi­neers and data sci­en­tists, to UX design­ers and prod­uct man­agers, have a chance to put forth inno­v­a­tive ideas that can evolve the way in which brands engage online and in the phys­i­cal world.

Lever­ag­ing the lat­est tech­nol­o­gy in areas like AI and mixed real­i­ty, sub­mis­sions are whit­tled down to a final set of 7 projects that are shared pub­licly. Over the years, Sneaks has become a core inno­va­tion engine for Adobe, deliv­er­ing capa­bil­i­ties like an AI assis­tant in ana­lyt­ics and per­son­al­ized web­site lay­outs, cho­sen by a machine. While these projects are not yet pub­licly avail­able, around 60 per­cent of Sneaks even­tu­al­ly make it into an Adobe product.

Project clothes swap

As more peo­ple are stay­ing home, online activ­i­ty is see­ing a surge. Online sales of appar­el for instance (dri­ven in part by pro­mo­tions) have increased 34 per­cent between March 12 and April 11 (per aggre­gat­ed data from Adobe Ana­lyt­ics). And while cat­e­gories like pants have fall­en 13 per­cent, more com­fort­able clothes in the pyja­ma cat­e­go­ry have risen 143 percent.

For years, appar­el brands have had to make them­selves more acces­si­ble beyond the phys­i­cal store. Ini­tial for­ays into eCom­merce have evolved into more sophis­ti­cat­ed uses of AI and data, as well as con­tent. This move was uni­lat­er­al, span­ning fast fash­ion to lux­u­ry retail­ers. Nike as an exam­ple, used a set of apps (e.g. SNKRS) to dri­ve affin­i­ty for the brand online. In the case of Pra­da, a 2019 col­lab­o­ra­tion with Adobe focused on using data and AI to get a bet­ter grasp of what con­sumers need­ed online.

The dig­i­tal store­front had become table stakes. For many now, it is a means of sur­vival. And once the world returns to a more nor­mal­ized state and con­sumer spend­ing resumes, we expect that brands will con­tin­ue invest­ing in their dig­i­tal strat­e­gy. We also believe that they will use tech­nol­o­gy in new ways, to be more effi­cient while meet­ing con­sumer demand.

We are show­cas­ing “Project Clothes Swap” in Adobe Expe­ri­ence Man­ag­er, to show how AI can take dif­fer­ent out­fits and move them around on dif­fer­ent mod­els online. To illus­trate how this works, imag­ine a brand that has a repos­i­to­ry of mod­els that have been pho­tographed in the past. When new styles come in, only the clothes have to be pho­tographed. Via Adobe Sen­sei, the AI will auto­mat­i­cal­ly form cloth­ing on a mod­el and deliv­er all the vari­a­tions a brand needs.

While we don’t expect phys­i­cal pho­to­shoots to go away, there will be instances where the time and cost sav­ings can be com­pelling – espe­cial­ly when the right tech­nol­o­gy is avail­able. The under­ly­ing AI in Project Clothes Swap, a patent­ed method called “SieveNet,” was cre­at­ed to remove dis­tor­tions that are char­ac­ter­is­tic of oth­er tech­niques. It aligns with Adobe’s long his­to­ry of work­ing with graph­ics and AI that has been trained to under­stand the nuances in com­po­si­tion, tex­tures and the like.

Project Clothes Swap can also play a help­ing hand when it comes to pro­mot­ing diver­si­ty in the images we see online. A recent study has shown that over 75 per­cent of female shop­pers pre­fer brands that fea­ture a vari­ety of eth­nic­i­ties in their ads. In the same way that one mod­el can show­case dif­fer­ent out­fits, the reverse is pos­si­ble as well.

As with many AI tech­nolo­gies, the use cas­es can be broad­ened and cus­tomized depend­ing on the brand. On the con­sumer side for instance, Project Clothes Swap can be lever­aged in sce­nar­ios where shop­pers want to upload their own pho­to and see how a piece of cloth­ing looks on them. The same goes for putting dif­fer­ent pieces of cloth­ing and acces­sories togeth­er one on mod­el, which the AI can help place accurately.