README.md

---
license: mit
tags:
  - electron-microscopy
  - 4D-STEM
  - materials-science
  - quantem
pretty_name: quantem-data
---

# quantem-data

**This page is in draft mode. We are still exploring the upload process
and the `meta.json` template.** Keys, the checker, and these instructions
will change as we try real pull requests. Use the current required keys
for new PRs, and expect the template to grow.

**This repo is currently used primarily by
[quantem.widget**](https://github.com/electronmicroscopy/quantem.widget)
(tutorial notebooks and `quantem.widget.datasets`). Other QuantEM packages
may use it later.

Public electron-microscopy data. MIT license. Downloads need no token.
This page is the upload and download protocol.

## Download

```python
from quantem.widget.datasets import show4dstem_gold

ds = show4dstem_gold(size="small")
```

Notebook: [Example Data](https://github.com/electronmicroscopy/quantem.widget/blob/main/docs/tutorials/download_data.ipynb).

## Add data (Hugging Face pull request)

Anyone with a Hugging Face account can add data. We follow a pull request
process. Do not use `quantem.widget.io.upload`.

Downloads need no token. Opening a pull request does. Create an account,
then a token at
[huggingface.co/settings/tokens](https://huggingface.co/settings/tokens).
Use a **Write** token (classic). A **Read** token can download; it cannot
open a PR. A fine-grained token also works if `--create-pr` is allowed on
a public dataset. You do not need collaborator access to this repo.

```bash
pip install huggingface_hub   # skip if `hf` is already on PATH
hf auth login
```

Paste the token. The prompt hides it. When asked `Add token as git
credential? [y/N]`, answer `n`. `hf upload` reads
`~/.cache/huggingface/token`. You do not need a git credential helper.
Ignore any hint to run `git config --global credential.helper store`.

```bash
hf auth whoami
```

That should print your username. If you already logged in with a Read
token, run `hf auth login --force` and paste a Write token. If
`--create-pr` returns 403 after a valid login, the token is Read or the
fine-grained scopes omit Community PRs; make a Write token and login
again. On SSH (no browser on that machine), paste a token. Do not use the
browser login unless the browser and `hf` run on the same machine. Do not
commit the token.

A PR is the data file(s) plus one `meta.json`. Example already on the hub
(2 files):

```text
widget-tutorials/show4dstem/gold-512-bin4/full/data.npy
widget-tutorials/show4dstem/gold-512-bin4/full/meta.json
```

One EMD: `gold_haadf.emd` + `meta.json` (2 files). A session folder: the
`.emd` files + `meta.json`.

Copy this `meta.json`. Required: `name`, `source`, `date`, `sample`,
`shape`, `dtype`, `permission`. Optional: `sampling`, `units`, optics.
Omit a key if unknown. Do not invent a number.

`shape` is `[row, col]` for an image, `[frame, row, col]` for a stack, or
`[scan_row, scan_col, det_row, det_col]` for 4D-STEM. `dtype` is the NumPy
name of the stored array (or of the EMD image). Examples already on this
repo:


| Data                      | `shape`              | `dtype`  |
| ------------------------- | -------------------- | -------- |
| 2D HAADF (`gold-haadf`)   | `[4096, 4096]`       | `uint16` |
| 4D-STEM (`gold-512-bin4`) | `[512, 512, 48, 48]` | `uint16` |
| One EMD HAADF             | `[4096, 4096]`       | `uint16` |


Processed float data uses `"float32"`. Read both values from the file
(`arr.shape`, `arr.dtype`, or the EMD image) and copy them into
`meta.json`.

```json
{
  "name": "gold_512_npy_bin4",
  "source": "Arina 4D-STEM, Nano Titan, session 20260423, gold_004_master.h5",
  "date": "2026-04-23",
  "sample": "gold nanoparticles",
  "shape": [512, 512, 48, 48],
  "dtype": "uint16",
  "permission": {
    "received": true,
    "from": ["Sangjoon Lee"],
    "collectors": ["Sangjoon Lee"],
    "date": "2026-04-23"
  },
  "sampling": [0.5, 0.5, 1.84, 1.84],
  "units": ["A", "A", "mrad", "mrad"],
  "voltage_kV": 300,
  "semiangle_mrad": 30,
  "camera_length_mm": 91
}
```

`source` is microscope, facility, and original file. `date` is acquisition
day (`YYYY-MM-DD`). `permission.received` must be `true`. `from` is who granted the share (one name or several). `collectors` is who
used the instrument (one name or several). `permission.date` is when that
share was granted.

```bash
python -m check_meta ./gold-512-bin4
hf upload bobleesj/quantem-data ./gold-512-bin4 \
  widget-tutorials/show4dstem/gold-512-bin4/full \
  --repo-type dataset --create-pr
```

Run `python -m check_meta` from a checkout of this dataset so the
`check_meta/` package is importable. `./gold-512-bin4` is your local
folder (`data.npy` + `meta.json`). The last `hf upload` path is the
folder on this repo. The hub path is
`widget-tutorials/<widget>/<name>/<size>/`. For a new ShowBragg set that
would be `widget-tutorials/showbragg/<name>/full`. Size is `small`,
`medium`, `large`, or `full`. Use `widget-tutorials/shared/` only when
more than one widget will read the same files.

`python -m check_meta` must print `ok` or do not open the PR. Resume a stopped
upload with `--revision refs/pr/N` (do not pass `--create-pr` again).
Merge is on the
[Community tab](https://huggingface.co/datasets/bobleesj/quantem-data/discussions).

### Coding agents

A coding agent can open the Community pull request. The human creates the
Write token and attests permission. The agent does not invent
`permission.*`, `sampling`, or optics. The agent does not run
`hf auth login` (that prompt is interactive). The agent does not write the
token into a repository file.

```bash
# human already ran `hf auth login` or exported HF_TOKEN (Write)
hf auth whoami
hf download bobleesj/quantem-data --repo-type dataset \
  --include "check_meta/**" --include "templates/meta.json" \
  --local-dir ./quantem-data-tools
export PYTHONPATH="$PWD/quantem-data-tools${PYTHONPATH:+:$PYTHONPATH}"
python -m check_meta ./gold-512-bin4
hf upload bobleesj/quantem-data ./gold-512-bin4 \
  widget-tutorials/show4dstem/gold-512-bin4/full \
  --repo-type dataset --create-pr \
  --commit-message "add show4dstem gold-512-bin4 full tutorial data"
```

`hf auth whoami` must print a username. If it fails or `--create-pr`
returns 403, stop and ask the human for a Write token (`export HF_TOKEN=...`).
If `python -m check_meta` is not `ok`, stop. After the upload, give the
human the Community URL. Do not merge.

Copyable widget prompt:
[Add tutorial data](https://github.com/electronmicroscopy/quantem.widget/blob/main/docs/agent-prompts.md#add-tutorial-data-hugging-face-pull-request).

If a notebook should call `quantem.widget.datasets`, open a GitHub PR after
this one merges:
[Contribute tutorial data](https://github.com/electronmicroscopy/quantem.widget/blob/main/docs/tutorials/contribute_data.md).

Colab: [Show4DSTEM](https://colab.research.google.com/gist/bobleesj/54864ce5b2a6f0a4fd5ae1e5d5719b45/show4dstem_colab.ipynb)
· [Workshop v1](https://colab.research.google.com/gist/bobleesj/a05a90185c6cddbb331342cae6d7e9c1/berk_workshop_v1.ipynb)

## Acquisition parameters

The 20260423 drift session optics are confirmed via the session HAADF EMD.
4D-STEM `scan_sampling` is an operator pattern from a sibling SSB session.


| dataset                          | voltage  | probe     | CL      | scan        | scan sampling  | det pitch    | mag         |
| -------------------------------- | -------- | --------- | ------- | ----------- | -------------- | ------------ | ----------- |
| `haadf/gold_haadf_npy`           | 300 kV ✓ | 30 mrad ✓ | 91 mm ✓ | 4096² image | 0.0186 nm/px   | n/a          | FOV 76.2 nm |
| `4dstem/gold_512_npy_bin8`       | 300 kV ✓ | 30 mrad ✓ | 91 mm ✓ | 512²        | 0.5 Å (op)     | 3.68 mrad/px | unknown     |
| `4dstem/gold_512_npy_bin4`       | 300 kV ✓ | 30 mrad ✓ | 91 mm ✓ | 512²        | 0.5 Å (op)     | 1.84 mrad/px | unknown     |
| `4dstem/gold_512`                | 300 kV ✓ | 30 mrad ✓ | 91 mm ✓ | 512²        | 0.5 Å (op)     | 0.46 mrad/px | unknown     |
| `4dstem/gold_30mrad1.3mx04`…`09` | 300 kV   | 30 mrad   | 91 mm   | smaller     | (session yaml) | 0.46 mrad/px | 1.3 Mx      |
| `haadf/gold_haadf.emd`           | 300 kV ✓ | 30 mrad ✓ | 91 mm ✓ | 4096² image | 0.0186 nm/px   | n/a          | FOV 76.2 nm |


✓ = confirmed via EMD/yaml. `(op)` = operator pattern, not file-certified.

## License

Files on this repo are published under **MIT**.

**You must have explicit written permission from the person who collected
the data before you open a pull request**, unless you collected the data
yourself. Put that receipt in `meta.json` (`permission.received`,
`permission.from`, `permission.collectors`, `permission.date`). If you
collected it yourself, set `from` and `collectors` to your name.

Do not open a pull request without that permission. `python -m check_meta`
must print `ok`.