README.md

---
license: cc-by-4.0
task_categories:
  - image-classification
  - feature-extraction
  - image-to-image
tags:
  - tactile
  - gelsight
  - gelsight-mini
  - robotics
  - tactile-sensing
  - pretraining
  - self-supervised
size_categories:
  - 100K<n<1M
pretty_name: GelSight Mini Pretrain
configs:
  - config_name: fota_labeled
    data_files:
      - split: train
        path: fota_labeled/train-*.parquet
      - split: val
        path: fota_labeled/val-*.parquet
  - config_name: fota_unlabeled
    data_files:
      - split: train
        path: fota_unlabeled/train-*.parquet
  - config_name: threedcal
    data_files:
      - split: train
        path: threedcal/train-*.parquet
  - config_name: feats
    data_files:
      - split: train
        path: feats/train-*.parquet
      - split: val
        path: feats/val-*.parquet
      - split: test
        path: feats/test-*.parquet
      - split: test_diff_sensor_new_gel
        path: feats/test_diff_sensor_new_gel-*.parquet
      - split: test_diff_sensor_old_gel
        path: feats/test_diff_sensor_old_gel-*.parquet
      - split: test_unknown_indenters
        path: feats/test_unknown_indenters-*.parquet
  - config_name: gelslam
    data_files:
      - split: train
        path: gelslam/train-*.parquet
      - split: recon
        path: gelslam/recon-*.parquet
  - config_name: tactile_tracking
    data_files:
      - split: train
        path: tactile_tracking/train-*.parquet
  - config_name: real_tactile_mnist
    data_files:
      - split: train
        path: real_tactile_mnist/train-*.parquet
      - split: test
        path: real_tactile_mnist/test-*.parquet
  - config_name: feelanyforce
    data_files:
      - split: train
        path: feelanyforce/train-*.parquet
  - config_name: sim_tactile_mnist
    data_files:
      - split: train
        path: sim_tactile_mnist/train-*.parquet
      - split: test
        path: sim_tactile_mnist/test-*.parquet
  - config_name: sim_starstruck
    data_files:
      - split: train
        path: sim_starstruck/train-*.parquet
      - split: test
        path: sim_starstruck/test-*.parquet
  - config_name: unit
    data_files:
      - split: train
        path: unit/train-*.parquet
  - config_name: tacquad
    data_files:
      - split: data_indoor
        path: tacquad/data_indoor-*.parquet
      - split: data_outdoor
        path: tacquad/data_outdoor-*.parquet
      - split: data_fine
        path: tacquad/data_fine-*.parquet
  - config_name: tvl
    data_files:
      - split: train
        path: tvl/train-*.parquet
---

# GelSight Mini Pretrain

![overview](assets/combined_overview.png)

A unified, parquet-native collection of **~1.1M [GelSight Mini](https://www.gelsight.com/gelsightmini/) tactile RGB frames** for self-supervised representation learning. **Ten** public datasets are aggregated under one schema, each filtered through a unified area+intensity contact rule with **per-domain thresholds** (I_min = 12 for real subsets, I_min = 10 for sim subsets, 1.5 % background-diversity keep rate) + per-capture phash dedupe:
- **~536K real-world frames** from 10 sources (FoTA labeled+unlabeled, 3DCal, FEATS, GelSLAM, TactileTracking, Real Tactile MNIST, FeelAnyForce, UniT, TacQuad)
- **400 K simulated frames** from 2 Mini-calibrated Taxim renders (Sim Tactile MNIST, Sim Starstruck)
- Plus a companion **CC-BY-NC extension** ([`yxma/gelsight-mini-pretrain-nc`](https://huggingface.co/datasets/yxma/gelsight-mini-pretrain-nc)) with another **~66 K** real frames from Meta Sparsh

Every row carries a `domain` column (`"real"` or `"sim"`) and a `markered` flag (gel has tracking dots?) so you can mix or filter freely.

![summary pies](assets/summary_pies.png)

➡️  **For a full per-subset breakdown** (intro, paper, license, processing recipe, sample grids, stats) see **[SOURCES.md](SOURCES.md)**.

📦  **Need more data and OK with non-commercial use?**  A companion
**CC-BY-NC extension** at [`yxma/gelsight-mini-pretrain-nc`](https://huggingface.co/datasets/yxma/gelsight-mini-pretrain-nc)
adds **~66 K** real frames from Meta's Sparsh dataset (flat / sharp /
sphere indenters with paired force ground truth). Same schema, same
pipeline, channel-order normalized. Load both and concatenate for the
largest pool. See the extension's
[sample images](https://huggingface.co/datasets/yxma/gelsight-mini-pretrain-nc#sample-images)
for a preview.

## TL;DR

```python
from datasets import load_dataset

# Largest subset, markerless, for VAE / MAE / contrastive pretraining
ds = load_dataset("yxma/gelsight-mini-pretrain", "fota_unlabeled", split="train")

# Pose-labelled subset for supervised fine-tuning
ds = load_dataset("yxma/gelsight-mini-pretrain", "fota_labeled",   split="train")

# Calibration sweep on a sphere indenter
ds = load_dataset("yxma/gelsight-mini-pretrain", "threedcal",      split="train")

# Markered (dotted) gel with force labels
ds = load_dataset("yxma/gelsight-mini-pretrain", "feats",          split="train")

img = ds[0]["image"]                       # PIL.Image (decoded on the fly)
meta = {k: ds[0][k] for k in ds.features if k != "image"}
```

## Why this exists

Public GelSight Mini data is scattered. Authors release each dataset with its own folder structure, file format (`.jpg` / `.png` / `.npy` pickled dicts / WebDataset tars), naming convention, and metadata layout. For most pretraining pipelines you end up writing four bespoke dataloaders. This release does the boring part once:

1. **Unifies the format**: every frame is a JPEG byte string in a parquet row, alongside a single shared metadata schema (poses, indenter labels, forces — all optional, all `null` where N/A).
2. **Re-encodes uniformly**: all images at JPEG quality 92, so file sizes are comparable across sources.
3. **Tags the gel variant**: a single `markered` bool tells you whether the gel has tracking dots — useful because dotted gels look visually very different from smooth ones, and mixing them confuses self-supervised objectives.
4. **Splits by source, not just train/val**: each upstream dataset is its own HF "config", so you can train on combinations of your choosing.

## How this dataset was built

The pipeline below was applied to **four public GelSight Mini datasets** (FoTA, py3DCal, FEATS, and — pending license — FeelAnyForce) to produce a single homogeneous parquet collection. Each step is a separate script in this repository's source tree; the four steps below are what changed *between* the raw upstream releases and what you `load_dataset` here.

![overview](assets/combined_overview.png)
<sub>*A 3-row overview across the three currently-released subsets. Each row is a different upstream dataset; every frame is a real GelSight Mini capture.*</sub>

**1. Source verification — "is this really GelSight Mini?"**
Each upstream release was checked against the Mini's known native modes (640×480 RGB, occasionally 320×240) and the dotted/dotless gel variants documented by GelSight Inc. We cross-referenced every dataset's `metadata.json`, paper, and folder naming convention to confirm the sensor model before inclusion. FAF is held back until licensing is clarified.

**2. Format unification — one schema, one image codec**
Upstream layouts are wildly different: FoTA ships **WebDataset tar shards**, py3DCal ships **loose PNGs in pose-named folders**, FEATS ships **`.npy` pickled dicts** containing the image plus a 32×24 depth grid plus forces. Each was decoded to a `numpy` array, re-encoded as **JPEG quality 92** for fair file-size comparison, and packed into the [unified schema](#unified-schema) below as a parquet row. Shards target ~2 GB each so they stream efficiently from HF Hub.

**3. Marker detection — fixing FoTA's mixed-gel surprise**
FoTA does **not** ship a markered/markerless label, but on visual inspection the captures are clearly mixed (some have ~80 visible tracking dots, others are smooth). We classify automatically: for each FoTA *capture* (one continuous press of an object), average ~50 evenly-spaced frames into a single mean image, then run dark-blob detection on it. Markered gels yield ≥10 well-sized dark dots in the mean; markerless gels yield <10 scattered noise blobs. Result: **36 of 124 captures are markered (all on the right finger), 88 are markerless**. The per-row `markered` column reflects this.

| Markered captures (top half) | Markerless captures (bottom half) |
|:---:|:---:|
| ![fota markered](assets/samples_40_fota_labeled_markered.png) | ![fota markerless](assets/samples_40_fota_labeled_markerless.png) |

**4. Empty-frame removal — cleaning FEATS**
FEATS' raw frames include captures where the indenter was hovering off the gel (no contact at all). These would teach a pretraining model nothing useful. We filtered them with `|f_z| < 0.5 N` (using FEATS's force-sensor ground truth), removing **5,302 frames** out of 22,013. We also added a `gel_variant` column distinguishing the two physical sensor setups used in FEATS (`black_dot` for the main markered gel, `different` for the second sensor used in `test_diff_sensor_new_gel`).

![feats](assets/samples_40_feats.png)
<sub>*FEATS samples after empty-frame removal — every frame now shows a real contact.*</sub>

**5. Tier-2 expansion: adding four more markerless Mini sources**
After the initial release a second pipeline pass added GelSLAM, TactileTracking, Real Tactile MNIST, and FeelAnyForce — four more public CC-BY/MIT-licensed GelSight Mini datasets. Each was processed through a shared backbone:

- **Adaptive subsampling** per source, capped at 200K kept frames so no single source dominates.
- **Validity filter** for video sources (GelSLAM, TactileTracking): a per-capture baseline is computed from the median of the first 10 frames; each subsequent frame is kept only if its central deformation from baseline exceeds ~4 grey levels. Up to 3% of kept frames are allowed below this threshold (variance).
- **Filter disabled** for already-curated sources (FeelAnyForce indentation stills, Real Tactile MNIST middle-of-touch frames) where every frame is by construction in-contact.
- **Perceptual-hash dedupe** within each capture (Hamming ≤ 4 on 8×8 DCT low-frequency hash) to drop near-identical adjacent frames from slow indentation videos.
- **Per-source frame budget** per dataset (see "Composition" above for kept counts; raw → kept summary in [§How this dataset was built](#how-this-dataset-was-built)).

Final result: **~865K frames** across 8 subsets, **~24 GB on disk**, one schema, one image codec, one `markered` flag.

The entire conversion script lives in this repo at [`scripts/make_parquet_v2.py`](scripts/make_parquet_v2.py). It implements the per-source decoders, the validity filter, the perceptual-hash dedupe, the 200K-per-source budget, and the parquet sharding. The CLI:

```bash
python make_parquet_v2.py probe   <subset>   # measure dynamism + empty fraction
python make_parquet_v2.py process <subset>   # full pipeline + write parquet shards
python make_parquet_v2.py stats              # row counts across all subsets
```

`<subset>` is one of `gelslam`, `tactile_tracking`, `real_tactile_mnist`, `feelanyforce`.

## Statistics at a glance

Composition across the 8 subsets, coloured by gel variant:

![composition](assets/composition.png)

Image resolution — the dataset retains both GelSight Mini native modes. Filter by `height`/`width` if you need a uniform-resolution pool:

![resolution](assets/resolution_distribution.png)

FEATS normal-force distribution and indenter-shape mix (the only subset with force labels):

![force](assets/force_distribution.png)

py3DCal probe-position coverage — confirms the dense calibration grid:

![threedcal coverage](assets/threedcal_coverage.png)

Real Tactile MNIST · digit-class balance (used as a sanity check that the upstream touch sampling was uniform across digits 0–9):

![rtm digits](assets/rtm_digit_distribution.png)

Per-channel pixel-value distribution comparing **kept** vs **rejected** frames for the three subsets that were filtered most recently (real_tactile_mnist with I_min=12; sim_tactile_mnist and sim_starstruck with I_min=10). The red/blue separation per channel confirms the filter discriminates contact-bearing frames from at-rest frames cleanly:

![pixel value distribution](assets/pixel_value_distribution.png)


<!-- BALANCE_SECTION -->
## Balance metrics

We report two complementary scores along four bucket axes — **domain**
(real/sim), **sensor_id** (13 distinct physical sensor configurations),
**object_id** (every unique object instance), and **gel_variant**
(markered/markerless):

- **Normalized Shannon entropy** `H̃ = H/log(B) ∈ [0,1]`. Higher = more
  uniform across buckets.
- **Effective Sample Size** `ESS = (Σn)²/Σn²`. Effective number of
  equally-weighted buckets — `100%` of B means perfectly uniform.

| Axis | B (buckets) | H̃ | ESS | ESS / B |
|---|---:|---:|---:|---:|
| domain | 2 | **0.95** | 2 | 94% |
| sensor | 13 | 0.69 | 5 | 35% |
| gel variant | 2 | 0.38 | 1 | 58% |
| object | **8,459** | **0.78** | 239 | 2.8% |
| 4-tuple bucket | 8,478 | 0.79 | 271 | 3.2% |

The dataset has **~63/37 real/sim** balance (H̃ = 0.95 along the domain
axis) and covers **8,459 unique object instances** across 13 physical
sensor configurations. The gel-variant axis is skewed toward markerless
(92.5% vs 7.5% markered) because most upstream sources only release
markerless captures.

See `assets/balance_report.json` for the full bucket histograms and
per-axis distributions.

## Composition

|       Subset      | Source dataset                          | Frames    | Gel        | Has labels                              |
|-------------------|------------------------------------------|----------:|------------|------------------------------------------|
| `fota_labeled`    | FoTA — *panda_warped* still captures    | **29,494** (66% markerless, 34% markered) | mixed¹ | end-effector x,y,z + quaternion |
| `fota_unlabeled`  | FoTA — same captures, video frames      | **266,761** train-only (re-sampled with looser dedupe; see ³) | mixed¹ | object name only            |
| `threedcal`       | py3DCal sphere indentation grid          | **36,270**| markerless | probe x, y, penetration depth (mm)       |
| `feats`           | FEATS indentation with force grids       | **16,711**| **markered** (two gel variants — see below) | indenter shape/size + contact forces |
| `gelslam`         | GelSLAM tactile SLAM tracking + reconstruction | **89,612** (28K tracking + 61K reconstruction) | markerless | episode + object name |
| `tactile_tracking`| TactileTracking (NormalFlow) 6DoF pose tracking | **1,605** | markerless | object + trial id |
| `real_tactile_mnist` | Real Tactile MNIST 3D-printed digit touches | **30,956** (26K train + 5K test; re-extracted from **video upstream** with peak-contact-per-touch picker, I_min=12) | markerless | digit class (0–9) + print id |
| `feelanyforce`    | FeelAnyForce force-controlled indentations | **50,997** | markerless² | object name |
| `sim_tactile_mnist` | **SIM** · Taxim-rendered Mini imagery of digit touches | **150,601** (102K train + 49K test; raw upstream + stride=2 + I_min=10) | markerless | digit class + episode |
| `sim_starstruck`  | **SIM** · Taxim-rendered Mini imagery of star objects | **166,104** (150K train + 16K test; raw upstream + stride=3 + I_min=10) | markerless | episode |
| `unit`            | UniT continuous 3D-pose tracking      | **11,340** | markerless | 3D-pose target (x,y,z,yaw) |
| `tacquad`         | TacQuad quad-sensor benchmark (Mini stream) | **12,195** (5K indoor + 4K outdoor + 3K fine, 181 objects) | markerless | object name + environment |
| `tvl`             | TVL Touch-Vision-Language paired tactile+RGB+caption | **209,795** (HCT + SSVTP subsets; multi-resolution: 640×480 + 320×240) | markerless | session id + paired RGB + GPT-4V caption (RGB+text not stored) |

¹ FoTA used **different gels on the two gripper fingers** for many of its captures. Approximately 36 of 124 captures use a markered gel on the right finger and a markerless gel on the left; the remaining 88 captures use markerless gels on both. The per-row `markered` column was set by averaging ~50 frames per capture and counting visible dark dots in the mean image (threshold ≥10 dots). Use it to filter:

² FeelAnyForce is colloquially described as a "markered Mini" dataset in some references, but visual inspection of the released tactile images confirms the gel surface is **smooth (markerless)**. The `markered` column reflects what is actually observed in the data.

³ `fota_unlabeled` was **uniformly stride-subsampled** from 516,523 → 66,761 frames (~12.9 % retention) and consolidated as **train-only** on 2026-05-18. Reason: only 11 objects across 60 captures means within-capture frames are near-duplicate gripper-trajectory snapshots, and the raw subset was ~70 % of the aggregated pretraining pool. Because this subset has only `obj_name` as a label (no pose / force / ground-truth contact) there is nothing to benchmark, so the prior train/val partition served no purpose and has been dropped. Every (object, init_pose, side) combination is still represented; the subset's share of the aggregated pool drops to ~7 %.

```python
ds = load_dataset("yxma/gelsight-mini-pretrain", "fota_labeled", split="train")
markerless = ds.filter(lambda r: not r["markered"])
markered   = ds.filter(lambda r: r["markered"])
```

## Sample images

### `fota_labeled` &nbsp;·&nbsp; 29,494 frames &nbsp;·&nbsp; mixed gel &nbsp;·&nbsp; +6DoF pose
One labeled still captured at every recorded end-effector pose along a Franka Panda trajectory pressing one of 13 household objects into the gel. The arc-shaped imprints are tactile signatures of objects (here, plier handles, clamps, knives, etc.). FoTA used **both markered and markerless gels** — use the `markered` column to filter.

Random sample (mixed):

![fota_labeled mixed](assets/samples_40_fota_labeled.png)

Per gel variant:

| markerless (66% of the data) | markered (34% of the data) |
|:---:|:---:|
| ![fota markerless](assets/samples_40_fota_labeled_markerless.png) | ![fota markered](assets/samples_40_fota_labeled_markered.png) |

### `threedcal` &nbsp;·&nbsp; 36,270 frames &nbsp;·&nbsp; markerless &nbsp;·&nbsp; +xyz pose
A motorised sphere indenter is pressed into the gel at **1,209 different (x, y) positions** at a fixed 3 mm depth. The bright spot moves as the probe walks across the sensor surface — useful for learning a calibrated position→appearance mapping.

![threedcal](assets/samples_40_threedcal.png)

### `feats` &nbsp;·&nbsp; 16,711 frames &nbsp;·&nbsp; **markered** &nbsp;·&nbsp; +force
Six indenter shapes (sphere, cuboid, cylinder, pyramid, cross, plus one "unknown" set of held-out probes), each pressed into a markered (dotted) GelSight Mini gel. Provides f_x, f_y, f_z forces and a 32×24 depth grid per image. Forces span from ~0 N (light touch) to **−73 N** (heavy normal compression).

![feats](assets/samples_40_feats.png)

> **⚠️ FEATS has two physical gel variants.** A new column `gel_variant` distinguishes them:
> - `"black_dot"` — the standard dotted Mini gel used for `train`, `val`, `test`, `test_unknown_indenters`, `test_diff_sensor_old_gel`.
> - `"different"` — a **second Mini sensor with a different gel** (smaller / dimmer / differently-coloured markers) used only in the `test_diff_sensor_new_gel` split. This split was designed by the FEATS authors to test cross-gel generalisation. Visually, marker dots are less prominent here.
>
> **No-contact frames removed.** The upstream FEATS dataset included ~5,300 frames where the indenter was hovering off the gel (|f_z| < 0.5 N). These have been filtered out here to give a cleaner pretraining set. Original counts (with the empty frames) were 22,013; current counts are 16,711.
>
> ![feats_diff_gel](assets/samples_feats_diff_gel.png)
>
> See `assets/samples_feats_by_split.png` for one example per (split, indenter) pair.

### `fota_unlabeled` &nbsp;·&nbsp; 66,761 frames &nbsp;·&nbsp; **train-only**
Visually identical to `fota_labeled` — same sensor, same objects, same captures — except these are the dense video frames between the labelled stills, with object name only (no pose). This is the bulk of the data and the primary target for self-supervised pretraining.

### `gelslam` &nbsp;·&nbsp; 60,982 frames &nbsp;·&nbsp; markerless &nbsp;·&nbsp; +per-episode 6DoF
The GelSLAM dataset of [Huang et al. 2025](https://arxiv.org/abs/2508.15990) — markerless GelSight Mini videos of an object being pressed into the gel and slid around. Two splits:
- `train` (27,763): the **tracking dataset** — 140 short episodes across 20 objects, each ~21 s long. Suitable for SLAM, pose-from-touch, and continuous pretraining.
- `recon` (33,219): the **reconstruction dataset** — 15 longer videos (1–30 min) of tactile scans across 15 objects (food, rocks, tool handles).

Per-frame `frame_idx` and per-row `episode` columns are populated. After empty-frame and dedupe filtering, **kept 21% of raw frames** (the majority of GelSLAM frames are pre/post-contact approach motion, which the validity filter removes).

![gelslam](assets/samples_40_gelslam.png)

### `tactile_tracking` &nbsp;·&nbsp; 1,143 frames &nbsp;·&nbsp; markerless &nbsp;·&nbsp; +per-trial 6DoF
The TactileTracking benchmark from the [NormalFlow paper](https://github.com/rpl-cmu/normalflow) (Huang et al. 2024) — 84 tracking trials across 12 objects. Aggressively filtered (empty-frame + perceptual-hash dedupe) because the trials are short and most consecutive frames are near-identical. Kept rate after filtering: 15%.

![tactile_tracking](assets/samples_40_tactile_tracking.png)

### `real_tactile_mnist` &nbsp;·&nbsp; 56,723 frames &nbsp;·&nbsp; markerless &nbsp;·&nbsp; +digit class
From the [Real Tactile MNIST benchmark](https://arxiv.org/abs/2506.06361) (Schneider et al. 2025) — **600 3D-printed MNIST digits**, each touched 256 times by a robot arm. The upstream release ships each touch as a short video clip; here we keep **one middle-frame per touch video** (near peak contact). After running the unified area+intensity contact filter with **I_min = 12** (calibrated for real-image noise floor; see `assets/pixel_value_distribution.png`), we retain 56,723 frames (28.4 % of the 200K raw frames; ~1.5 % background-diversity keep rate). The `digit_class` column gives the digit 0–9; the `episode` column gives the print id (which of the 600 physical digits was touched).

![real_tactile_mnist](assets/samples_40_real_tactile_mnist.png)

### `feelanyforce` &nbsp;·&nbsp; 50,997 frames &nbsp;·&nbsp; markerless &nbsp;·&nbsp; +per-indentation object
The [FeelAnyForce dataset](https://huggingface.co/datasets/amirsh1376/FeelAnyForce) (Sharei et al. 2024) of robotically-controlled indentations against 42 unique objects (cylinders, cubes, spheres, fruits, household items). Aggressively dedupe-filtered (50% of raw frames dropped) because each indentation is a slow press-and-hold, so adjacent frames are visually near-identical. The schema retains the `obj_name` and `episode` columns; the upstream force labels (in `TacForce_train_set.csv` etc.) are not currently joined in — see the [upstream release](https://huggingface.co/datasets/amirsh1376/FeelAnyForce) for forces.

![feelanyforce](assets/samples_40_feelanyforce.png)

## Useful statistics

### `fota_labeled` — pose coverage
- **13** distinct contact objects · **5** initial poses · **15,148 unique (object, pose, side) captures**
- Frames split equally between left / right gripper finger: **14,747 / 14,747**
- End-effector range: `x ∈ [−25.7, 129.1] mm` · `y ∈ [−137.3, 137.3] mm` · `z ∈ [−38.6, 38.6] mm`
- Top objects by frame count:

|object|frames|
|---|---:|
|tapemeasure | 4,800 |
|whiteclamped| 3,600 |
|blackclamp  | 3,016 |
|blackclampclosed | 2,772 |
|key1 | 2,400 |
|plierhandle | 2,400 |
|foldingknife | 2,156 |
|wrench | 2,134 |
|blackclampedclosed | 1,922 |
|printedstrawberry | 1,800 |

### `threedcal` — calibration grid
- All 36,270 frames are sphere indentations at **z = 3 mm penetration**
- `x ∈ [0, 19] mm` · `y ∈ [0, 15] mm` ·  **1,209 unique (x,y) grid positions**
- ~30 repeated frames per position (lighting noise / repeat trials), useful for variance estimation

### `feats` — indenter & force coverage

| Shape | Sizes (mm) | Total frames |
|---|---|---:|
| cuboid    | 2, 7, 10, 12, 15, 20 | 4,683 |
| sphere    | 10, 15, 20           | 2,883 |
| cylinder  | 8, 10                | 1,291 |
| cross     | 15                   | 894   |
| pyramid   | 10                   | 854   |
| *(unannotated)* | —              | 11,408 |

- Normal force range: `f_z ∈ [−73.3, 0.0] N` (mean **−9.30 N**, std **10.66 N**)
- Shear force range: `f_x ∈ [−4.86, 4.86] N`, `f_y ∈ [−5.89, 5.87] N`
- Six labeled **splits** including out-of-distribution tests:
  - `train` (15,670), `val` (921), `test` (1,845) — in-distribution
  - `test_diff_sensor_new_gel` (395) — new physical gel
  - `test_diff_sensor_old_gel` (393) — different sensor unit
  - `test_unknown_indenters` (2,789) — held-out probe shapes

### `fota_unlabeled` — characterisation
- **11 distinct objects** (subset of `fota_labeled`'s 13)
- **60 captures** (11 objects × ~5 init-poses × 2 gripper sides)
- Roughly balanced left / right (~50/50)
- ~1000 frames per capture after stride-subsampling³

## Unified schema

Every row, regardless of source, has the same columns. Optional fields are `null` when not applicable.

| Column | Type | Description |
|---|---|---|
| `image`          | image (binary)  | tactile RGB frame, JPEG bytes; auto-decoded by `datasets` to `PIL.Image` |
| `image_format`   | string  | always `"jpeg"` |
| `source`         | string  | one of `"fota_labeled"` / `"fota_unlabeled"` / `"3dcal"` / `"feats"` |
| `markered`       | bool    | does the gel have tracking dots? |
| `capture`        | string  | per-source capture / scene / probe identifier |
| `split`          | string  | dataset split (e.g. `train`, `val`, `test_unknown_indenters`, …) |
| `height`, `width`| int32   | image dimensions in pixels |
| `obj_name`       | string  | (FoTA) which object was pressed |
| `init_pose`      | int32   | (FoTA) initial-pose index for this capture |
| `side`           | string  | (FoTA) `"left"` or `"right"` gripper finger |
| `x_mm`, `y_mm`, `z_mm` | float32 | probe / end-effector position |
| `quat_x..w`      | float32 | (FoTA labeled) end-effector orientation |
| `indenter`       | string  | (FEATS) probe shape, e.g. `"sphere"` |
| `indenter_param` | string  | (FEATS) probe size in mm |
| `f_x`, `f_y`, `f_z` | float32 | (FEATS) total force on probe (N) |
| `grid_z_max`, `grid_z_mean` | float32 | (FEATS) summary of per-pixel depth grid |
| `gel_variant`    | string  | (FEATS only) `"black_dot"` (standard markered gel) or `"different"` (the second sensor / different gel used in `test_diff_sensor_new_gel`) |
| `domain`         | string  | `"real"` for real-world captures or `"sim"` for Taxim-rendered Mini imagery |

## Real vs simulated data

Every row carries a `domain` column. The 758 K real-world frames span 8
upstream datasets capturing physical robot–object contact on a real
GelSight Mini sensor; the 400 K simulated frames come from the
**Mini-calibrated Taxim renderer** of Schneider et al.

```python
# Real-only markerless pool (largest pretraining target)
real_markerless = concatenate_datasets([
    load_dataset("yxma/gelsight-mini-pretrain", c, split="train"
                ).filter(lambda r: r["domain"] == "real" and not r["markered"])
    for c in ["fota_unlabeled", "threedcal", "gelslam", "tactile_tracking",
              "real_tactile_mnist", "feelanyforce"]
])

# Sim pool — useful for sim-to-real transfer or as augmentation
sim_pool = concatenate_datasets([
    load_dataset("yxma/gelsight-mini-pretrain", c, split="train")
    for c in ["sim_tactile_mnist", "sim_starstruck"]
])
```

The two sim subsets are from the **same authors as `real_tactile_mnist`**
(Schneider et al. 2025) and use the same Mini sensor model. Sim frames
are noise-free, so the contact filter is applied with a lower threshold
**I_min = 10** (vs I_min = 12 / 15 for real subsets), yielding ~93 %
retention for `sim_tactile_mnist` and ~97 % for `sim_starstruck`. See
`assets/pixel_value_distribution.png` for the per-channel pixel-value
histograms showing how the filter discriminates contact vs at-rest
frames in each subset.

## Recommended uses

- **Self-supervised pretraining** (VAE / MAE / SimCLR / DINO):
  use `fota_unlabeled + fota_labeled + threedcal` (markerless, **~133K frames**).
  Hold `feats` out since its dotted-gel appearance will dominate the
  reconstruction objective if mixed in.
- **Force / shear regression**: fine-tune on `feats` with its `f_{x,y,z}` labels.
- **Pose estimation**: fine-tune on `fota_labeled` (xyz + quat) or `threedcal`
  (xy + depth).
- **Marker invariance studies**: train markerless ↔ test on `feats`, or use a
  marker-mask augmentation.

## Provenance and citation

This dataset is a re-packaging of existing public datasets — please cite the
upstream sources if you use the data:

- **FoTA** — *Foundation Tactile*, Zhao et al., 2024.
  [HF dataset](https://huggingface.co/datasets/alanz-mit/FoundationTactile),
  [GitHub](https://github.com/alanzjl/t3),
  [arXiv:2406.13640](https://arxiv.org/abs/2406.13640) · *MIT License*.
- **3D Cal (py3DCal)** — Kota, Shah, Colgate, Reardon (2025).
  [Zenodo 18462608](https://zenodo.org/records/18462608) · *CC-BY-4.0*.
- **FEATS** — Helmut (2025).
  [HF dataset](https://huggingface.co/datasets/erikhelmut/FEATS) · *MIT License*.
- **GelSLAM** — Huang et al., 2025.
  [HF dataset](https://huggingface.co/datasets/joehjhuang/GelSLAM_dataset),
  [GitHub](https://github.com/rpl-cmu/gelslam),
  [arXiv:2508.15990](https://arxiv.org/abs/2508.15990) · *MIT License*.
- **TactileTracking / NormalFlow** — Huang, Kaess, Yuan (2024).
  [HF dataset](https://huggingface.co/datasets/joehjhuang/TactileTracking),
  [GitHub](https://github.com/rpl-cmu/normalflow),
  IEEE RA-L 2024 · *MIT License*.
- **Real Tactile MNIST** — Schneider et al., 2025.
  [HF dataset family](https://huggingface.co/TimSchneider42),
  [GitHub](https://github.com/TimSchneider42/tactile-mnist),
  [arXiv:2506.06361](https://arxiv.org/abs/2506.06361) · *CC-BY-2.0*.
- **FeelAnyForce** — Sharei et al., 2024.
  [HF dataset](https://huggingface.co/datasets/amirsh1376/FeelAnyForce) · *CC-BY-4.0*.
- **Sim Tactile MNIST / Sim Starstruck** — Schneider et al., 2025 (same authors as Real Tactile MNIST).
  [`tactile-mnist-touch-syn-single-t32-320x240`](https://huggingface.co/datasets/TimSchneider42/tactile-mnist-touch-syn-single-t32-320x240),
  [`tactile-mnist-touch-starstruck-syn-single-t32-320x240`](https://huggingface.co/datasets/TimSchneider42/tactile-mnist-touch-starstruck-syn-single-t32-320x240).
  Mini-calibrated Taxim renderer. *CC-BY-2.0*.
- **Taxim** simulator (used by the above sim sources) — Si & Yuan, 2022.
  [GitHub](https://github.com/Robo-Touch/Taxim),
  [arXiv:2109.04027](https://arxiv.org/abs/2109.04027).

Conversion details:
- All images are re-encoded to JPEG at quality 92. Original PNGs in
  `threedcal` are decoded losslessly and re-encoded; FEATS `.npy` dicts have
  their `gs_img` array extracted and saved as JPEG; FoTA WebDataset shards
  are unpacked and re-encoded. Metadata is preserved verbatim.
- *FeelAnyForce* (Sharei et al., 2024) ~200K markerless frames may be added
  in a future revision pending license verification with the upstream authors.

## License

This aggregated release is released under **CC-BY-4.0** (the most restrictive
license among the included sources). Cite the component datasets above.

## Acknowledgments

Thanks to the FoTA, py3DCal, and FEATS authors for releasing their data
publicly.