I would like to download images from Tinder to use with an image classification app. (Tensorflow with Imagenet to be precise.) Is it illegal to download those images for that purpose?
Also, can I post my findings and/or write a publication based on them?
Tinder's Terms of Service (TOS) is pretty clear:
Therefore, you agree not to:
• use the Service or any content contained in the Service for any commercial purposes without our written consent.
• copy, modify, transmit, create any derivative works from, make use of, or reproduce in any way any copyrighted material, images, trademarks, trade names, service marks, or other intellectual property, content or proprietary information accessible through the Service without Tinder’s prior written consent.
• use any robot, bot, spider, crawler, scraper, site search/retrieval application, proxy or other manual or automatic device, method or process to access, retrieve, index, “data mine,” or in any way reproduce or circumvent the navigational structure or presentation of the Service or its contents.
• modify, adapt, sublicense, translate, sell, reverse engineer, decipher, decompile or otherwise disassemble any portion of the Service, or cause others to do so.
• use or develop any third-party applications that interact with the Service or other users’ Content or information without our written consent.
The Company may investigate and take any available legal action in response to illegal and/ or unauthorized uses of the Service, including termination of your account.
If you ignored that and did your machine learning research with images anyway, you're breaking the TOS, and to be honest, you may not be caught. But if you published your research, Tinder could demand you engage in arbitration or sue you on the premise that your research proves you violated their TOS.
If your jurisdiction is outside of the US, other laws may come into play, but it's safe to assume you could still be in legal jeopardy from Tinder.
You don't make it clear whether you plan to manually download some number of images, or use some sort of script or app to do the downloading in bulk. As the TOS quoted in BlueDogRanch's answer makes clear, using any sort of automation is forbidden by Tinder's TOS.
The TOS also forbids anyone to:
copy, modify, transmit, create any derivative works from, make use of, or reproduce in any way any copyrighted material, ... (emphasis added)
This might well cover even manual downloading and separately using the images in such an app -- unless you seek and obtain specific permission from Tinder -- which perhaps they would give.
The question mentions using:
Tinder or similar platforms
I would suggest seeking a "similar platform" with a different TOS.
Then there is the question of actual copyright infringement, over and above any TOS violations. If the images were licensed under a free license, such as most of the Creative Commons licenses, this would not be an issue. But if they are not, you would need to depend on Fair Use (if in the US, you didn't state your jurisdiction). The answer by ser6726 indicated grounds for thinking this would work, but details matter in fair use judgements, and you would be wise to seek the advice of a lawyer with copyright expertise to whom you can provide more specifics, or else to stick to images with free licenses.
Downloading photos will indeed be against the ToS, however note that:
Tinder can stop serving you if you violate the ToS. In order to sue you, they will have to demonstrate the damage you did to them. Constantly and measurably overloading a server with automated requests would be one type of damage. Arguably, this will be hard to from a single personal computer, unless your requests are computationally expensive. Downloading photos is unlikely to cause this, but actual cases claiming such damages do exist, e.g. EBay v. Bidder's Edge.
Illegally distributing a derivative work could constitute copyright infringement, which comes with statutory damages - the plaintiff will be able to sue you for a fixed amount of money without having to prove how much damage you actually caused. Depending on circumstances, you might be able to claim that your actions constitute fair use. In any case, it is recommended not to publish the data set itself until you clear its status with an actual lawyer. Simply claiming to have the data set without publishing it (e.g. in a scientific paper) significantly reduces the chances of success of a potential lawsuit.
There are two issues to consider in figuring out the legality of such copying. First, is it a violation of the TOS for the website. Second, is it excused under "fair use"? To answer the first question, you have to read the TOS for a particular web page / service, looking for prohibitions. It is possible but unlikely that it is clearly prohibited by the hosting service (largely because their concerns are about copyright infringement, not blocking research – however, your means of downloading the images could be contrary to the TOS, so check that). As you read through the info on fair use, you'll see that there is a balancing act that the courts perform. Briefly, the factors are the commercial / nonprofit educational nature of the copying, the nature of the original, the amount copied, and the effect of the use on the market. Suppose you are writing a program that looks for pictures of dogs and you need a supply of dog pictures to see if the program works: this is your PhD thesis topic. You are covered w.r.t. nature of copying. You are also covered w.r.t. effect on market (presumably: you are not redistributing the images and there is probably no royalty attached to looking at an uploaded photo).
It's hard to say about the "nature of original", in that "highly creative" works are more protected than "minimally creative" works, but probably that one goes in the "not fair use" column. And you presumably are copying the whole thing. There is a tie-breaker, "transformativeness" (being transformative favors fair use). Simply copying and making a minor tweak is not very transformative; copying an image and transforming the image into a number like "75% chance that this is a dog" is very transformative.
There is no official weighting of the four factors, however, outcomes in court cases indicate that the copying necessary to carry out academic research is deemed "fair use". The doctrine was created to allow commentary and research, and the courts strongly favor those kinds of copying.
