Data download links (HT-LIF24 3 ms/5 ms, CARE3D, HHMI-D25) return HTTP 503; availability of pretrained checkpoints

[baoyi-A]

We're reproducing several Table 1 tasks to benchmark against our own method, and have hit two access issues we couldn't resolve from the public README alone.

1. Dataset download links return HTTP 503
The following download.fht.org links (referenced in the README, incl. around README lines 164–165 for the HT-LIF24 exposures) consistently return HTTP 503 for us, retried across several days (2026-07-03/04/05) and from two networks:

• HT-LIF24 3 ms and 5 ms exposure archives (Tasks XIV, XXV, XXVI, XXIX)
• CARE3D (CBG-Z18 / CBG-N18) — Tasks XXI, XXII
• HHMI-D25 archives (8-bit / 16-bit / binned / noise variants) — Tasks XXIII, XXXI–XXXVI

The 500 ms, 2 ms, and 20 ms HT-LIF24 data downloaded fine; only the above return 503. Are there working mirrors, or is the host temporarily down?

2. Pretrained checkpoints
Are per-dataset/per-task pretrained checkpoints available anywhere (the exact models behind the Table 1 numbers)? We already have the HT-LIF24 500 ms checkpoint. Having the released models would let us compare against your exact runs under an identical pipeline rather than only against the paper numbers.

3. (Secondary) P23A / ExpMicroscopyV1 2D config
The repo has configs/data/exp_microscopy_v1.py but we couldn't find a matching 2D dataset loader / parameters module for P23A. Is there a 2D config+params for ExpMicroscopyV1, or was P23A only run in 3D?

Environment: careamics 0.0.9, microsplit_reproducibility (current main), torch 2.5.1+cu124, Python 3.9, Linux.

Thanks!

[CatEek]

HI @baoyi-A
Thank you for your interest. We currently have an ongoing server maintenance, so both datasets and checkpoints are not available unfortunately, We're sorry for the inconvenience.
About point 3 I'm not sure I follow, could you please clarify

[baoyi-A]

Thanks, we'll grab the data and checkpoints once it's back up.

For point 3: in current main I only see per-dataset modules for the HT_* datasets (HT_H23B, HT_H24, HT_LIF24, and HT_T24 has just the parameters file), nothing for P23A/P23B, Pavia, Chicago-Sch23, or CARE3D. So my question is really about those — especially P23A (ExpMicroscopyV1): are configs/loaders for them planned, or is the idea to build on the custom_dataset_2D/3D templates? I'd seen an exp_microscopy_v1 config in an older version, which is what I was originally referring to.

Thanks!

[jdeschamps]

Hi @baoyi-A,

There is currently no plan to implement any additional experiment here. We are still refactoring MicroSplit in order to have a more usable and modular version integrated into CAREamics (CAREamics/careamics#972).

We can search for the checkpoints as soon as we will have access back to our servers.

Regarding HT-P23A/B, as in #27, best would be to start from a 3D example (they are 3D experiments in the manuscript). As soon as servers are back we can dig into channels and noise models data so you can create your own configuration.