How to enable facial recognition in Feiniu Album? My home tested 40,000 photos—can the N100 really hold up?

during the New Year, my mom wanted a photo of her holding the baby, saying it was taken three years ago. I spent almost an hour flipping through the Feiniu album, over 40,000 photos, my eyes were spinning. The next day after I finished flipping through it, I went to unlock the facial recognition—something I should have opened long ago, but I kept delaying it because I heard it would take a long time and was nervous. After this real-world test, I'll explain all at once: how long it will run, whether it's accurate, and whether the N100 can handle it.

clarify first: the recognition runs inside your own machine

Many people wonder when they hear facial recognition: Where did my family's faces go? This matter needs to be clarified first. AI albums like Baidu Netdisk and iCloud send photos to the server's server for calculation, and only after recognition do they get the results. Feiniu is different; the recognition is done by the NAS's own CPU, and the calculated character database and feature data are all stored on your local hard drive. Even when offline, it can still be recognized, and photos can be taken without leaving the house.

I specifically checked the data once and after enabling recognition, the NAS had no external photo data and only uploaded my phone's album auto-backing up. So there's no need to worry about privacy; the real concern is the other side—your machine's performance might not be enough to keep up with its messaging. I'll explain that in detail below.

: Don't miss a switch. Before activating it, set two things

the entry point is in the Album app settings. Open your album, go to Settings, find the "Face Recognition" switch, and turn on the "Smart Recognition" (some versions call it Scene Recognition) as well. These two are separate switches; just turning on the face switch is a loss. Smart recognition is actually more practical than facial recognition.

my device is version 0.9.x, and the settings are right here. Feiniu updates quite frequently, and the menu occasionally shifts. If you can't find it, just search for the word "face" in the system—it's more reliable than my description. One more reminder: the older versions don't have facial recognition at all—upgrade first and then look for it. Before

flip the switch, two things are set first. First, time and budget. Once activated, it doesn't work immediately; it scans all your photos to count features. My family took over 40,000 photos in a day and a half, during which you can't check faces. Second, pick a gap period; during recognition periods, the machine load is high, so don't turn it on while you're using the NAS for real work.

N100 40,000 photos: report cards and side effects you want to know

My laptop is an N100 mini PC, 16GB RAM, Feiniu, and dual hard drives. The photos were saved over the past few years by automatic phone backups—over 41,000 images, including screenshots and messy images.

report card: After running everything, it took more than a day and a half. The four CPU cores in the middle are occupied 70% to 80% year-round, and in Task Manager, several processes related to recognition are taking turns eating. As for memory, the recognition service itself consumes more than 1GB. If you add systems like OS and Docker, machines 4GB memory really shouldn't be forced to . When I had 4GB, just running the photo album index was a struggle, and I only upgraded to 16GB later. If you want to upgrade memory, check NAS the about memory upgrades. I tested the differences between 4GB, 8GB, and 16GB levels.

three side effects, all of which I have experienced. First, the Jellyfin transcoding was noticeably laggy over the past couple of days. The N100 used to double as a video library, but Docker the Jellyfin installation , I mentioned that the N100 transcoding is sufficient but not wealthy, and once it runs for recognition, it gets even more crowded. Second, hard drive sleep is completely ruined. The recognition process keeps reading and writing to the library. The sleeping strategies you set up before are basically nonexistent for those two days. Not hibernation itself doesn't harm the machine, but it does cost more electricity. If you mind who steals the sleep, read on hard drive sleep troubleshooting. Indexing tasks are one of the main sources of hidden disk writing. Third, the moment you back up new photos on your phone is half a beat slower than usual, as the recognition is stealing resources, which is acceptable.

So my suggestion is to open on Friday night and let it run slowly throughout the weekend, basically ready before Monday work.

accurate: Adults have steady faces, children are the hardest-hit areas

After the recognition run, get to the main topic—is it really up to standard? First, the stability part: the accuracy of adult faces exceeds my expectations, over 90%. I know my husband's photos perfectly—even my dad's black-and-white re-photos from thirty years ago are credited to him, and the two sets of photos of my father-in-law with glasses and without glasses have been merged together. Profiles and silhouettes can't be easily brushed off.

the hardest hit are children. From birth to age three, the baby's face changes year by year. The machine categorizes them by 'look-alike', resulting in a pile of one-year-olds and a pile of two-year-olds. The same face is split into three or four people. What's even more infuriating are the two kids. My eldest and second are two years apart. They looked alike as kids, and the machine would often stuff photos of the younger brother into the older brother's character card. in the comparing Synology and Feiniu, I mentioned that Feiniu's facial recognition is a notch worse than Synology's. The main difference is this: Synology Photos can also be removed from children, but there are fewer mismatches.

luckily, there's a way to save it—let's merge manually. Open the character card you admitted your mistake, long-press the photo and select 'This is not the person,' then pick the right person from the list. You can check dozens of photos at once. I spent an entire night sorting out the photos of the two kids. You can change names on character cards. I gave my two kids nicknames, and my mom can search for 'brother' in the photo album app and it appears. For twin families, I advise you not to argue—human eyes are always different, so let the machine go with the flow.

intelligent recognition and image search, honestly, it's even more useful than facial recognition

and the smart recognition I just turned on casually, turns out to be the best value for me. It categorizes photos by type: food, plants, buildings, screenshots, documents—each belonging to its own category. Last week, I found a photo of a medical checkup report from half a year ago, flipped through the screenshot category, and found it in two minutes. In the past, I had to recall it by month and grid by month.

search box has become smarter too. Getting a 'hotpot' can help you recover all the hotpot you've had in the past two years, and issuing an 'invoice' can list the reimbursement vouchers you've saved. This accuracy is a bit worse than a face, and there are often inexplicable accidents—my cat's sleeping posture is classified as a 'stuffed toy,' but it's more than useful for finding things.

two pitfalls, and who is worth digging

pit to talk about. First, if the power goes out halfway through a run, the progress won't be fully redone, but there will be a small repeat run. My family once lost power once, which took about two extra hours. So during those two days of running the recognition test, don't touch the power supply. Second, after changing machines and reinstalling the system, the character library doesn't follow the photos; the new machine requires a re-recognition run. Fortunately, the photos themselves are still there. The data transfer method the hard drive migration article detailed, so recognition is treated as paying a new tuition fee.

order for you: back up all the photos first, upgrade to the new version, turn on both switches on Friday night, let them run on weekends, spend an hour on Monday night merging and renaming the child's face, and after that, finding photos will be a second away.