Validity of the project.

[Mathew005]

Could someone share proof of the execution results claimed in this project under a real-world setup, including the actual hardware used? From what I can see, the repository mainly contains hypothetical claims across various .md files, and I haven’t been able to reproduce the reported results even with the hardware specifications mentioned in the repo.

[npdias]

I was looking at some Esp32's I've got in a drawer to test myself. What did you try doing? What results did you get?

[LH-star]

hello，did you get some result

[xiaofuchen]

我全程试了，不能实现，里面的算法几乎可以理解成是随机数，没有实际应用意义。

[thematica]

I bought 3x ESP32-S3 to try this project, but I don't think this works as advertised

observatory.html didn't show real ESP32 data, only demo animation/data even I point to real data via websocket.

i put off 1 node and the data didn't even changed much

oh well, there goes my $30

[jeblove]

I bought 4 ESP32-S3 boards specifically to try to reproduce this project.

After testing the setup, I found that with ESP32 connected, /api/v1/sensing/latest seems to update only once after reset, and then the values remain fixed without further changes. Because of that, I cannot verify that the data is actually real-time or genuinely derived from live sensing.

The same behavior happens with /api/v1/pose/current: only the timestamp keeps updating, while the other values remain unchanged.

As for the Real-time CSI Data shown under /ui/index.html → Hardware, I disconnected power from all ESP32 boards, but those values kept changing😄. Is that simulated data?

Also, I have not seen any evidence that anyone other than the author has successfully reproduced the full setup.

[xiaofuchen]

#568

[atan02]

I have three units and I haven't been able to get it to work remotely close to their claims. Problem 1, the ESP32-S3 are very sensitive to placement. The CSI can widely fluctuate making it unreliable for consistent detection. Problem 2, training data is non-existent to determine presence, movement, heart rate etc. I think you have to build up your own. I may switch over to Espressif's repo to see if they are any better. Disappointed!!!

[sbrown89]

The underlying Wi-Fi CSI technology is legitimate, and there appears to be substantial real engineering work in this repository. What I am having difficulty reconciling is the breadth of the README with the limits described in the project’s own proof documents and model cards.

For example, the current 2.4 GHz encoder documentation says it was trained from one room, one capture, and two nodes, and clarifies that its temporal-triplet metric is not downstream presence or vital-sign accuracy. The pose result is much stronger on the matched MM-Fi benchmark, but the published generalization results also indicate a significant drop when moving to an unfamiliar environment without calibration.

That does not make the project fake. It suggests that presence, motion, pose, vital signs, counting, and medical-adjacent features are at substantially different maturity levels.

My two cents is that RuView is promising Wi-Fi sensing research rather than a broadly validated, deployable "see through walls" system.
Are there independent hardware reproductions showing the pose or vital-sign features working in previously unseen rooms with previously unseen people?