UniT: re-ingest with contact filter (11,340 → 387 kept); fix gel_variant to markered (gel has visible dot pattern). Update docs: remove TVL (was DIGIT not Mini), add to Investigated. Grand total: 852,915 frames.

README.md

@@ -88,10 +88,6 @@ configs:
         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
@@ -266,9 +262,8 @@ per-axis distributions.
 | `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) |
+| `unit`            | UniT continuous 3D-pose tracking      | **387** (heavily filtered; only ~3% of 11,340 raw frames have actual LED-on contact) | **markered** | 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:
 

SOURCES.md

@@ -540,7 +540,13 @@ bearing.
 
 | Subset | Frames | Resolution | Markered / Markerless | Splits |
 |--------|-------:|------------|----------------------:|--------|
-| `unit` | 11,340 | 320 × 240  | 0 / 11,340            | train 11,340 |
+| `unit` |    387 | 320 × 240  | 387 / 0               | train 387 |
+
+After re-ingesting with the contact filter (initially we kept all 11,340
+frames assuming continuous contact, but 86% turned out to be near-black
+LED-off frames), only **387 frames** show actual gel contact. The visible
+frames clearly show **markered gel** (regular dot pattern), so this
+subset goes in the markered pool.
 
 **40 random samples:**
 
@@ -591,53 +597,6 @@ real-world objects) in the entire dataset.
 
 ---
 
-## 12 · TVL — Touch-Vision-Language paired tactile + RGB + caption (`tvl`)
-
-**Intro.** TVL (Fu et al., ICML 2024) is a multimodal alignment dataset
-with ~44K paired vision-touch examples plus English captions (10%
-human-annotated, 90% GPT-4V pseudo-labels). The underlying tactile
-sensor is GelSight Mini and the dataset has two subsets:
-- **HCT** (hand-collected touches): 3 data folders of timestamped touch
-  trajectories with separate `tactile/` + `vision/` subfolders per
-  touch.
-- **SSVTP**: a smaller subset with `images_tac/` + `images_rgb/` +
-  `text/` for human-captioned alignment.
-
-We ingest **only the tactile images**. Captions and paired RGB are
-preserved upstream but not stored in this aggregate.
-
-**Source release.**
-- 🤗 [`mlfu7/Touch-Vision-Language-Dataset`](https://huggingface.co/datasets/mlfu7/Touch-Vision-Language-Dataset)
-  (original, has a known tactile/image folder-swap bug)
-- 🤗 [`yoorhim/TVL-revise`](https://huggingface.co/datasets/yoorhim/TVL-revise)
-  (corrected fork — **what we use**)
-- 📜 License: research/academic use only (paper-stated; no explicit
-  CC license in HF metadata)
-
-**Original format.** 8-part sharded zip (~75 GB total). After combining
-with `zip -s 0` and unzipping, ~410K JPEG files split across HCT
-(640×480) and SSVTP (240×320) — **two distinct tactile resolutions**.
-Pipeline handles per-resolution baselines.
-
-**How we processed it.** Walked HCT + SSVTP tactile-only directories,
-applied area+intensity filter (I_min=12) **per resolution** (separate
-gel-at-rest baselines for 640×480 vs 240×320). Channel-order
-auto-normalized.
-
-**Stats after processing.**
-
-| Subset | Frames | Resolution | Markered / Markerless | Splits |
-|--------|-------:|------------|----------------------:|--------|
-| `tvl`  | 209,795 | mixed (640×480 HCT + 240×320 SSVTP) | 0 / 209,795 | train 209,795 |
-
-The dataset is dominated by the HCT subset (~200K@640×480) with a
-smaller SSVTP contribution (~10K@240×320).
-
-**40 random samples:**
-
-![tvl](assets/samples_40_tvl.png)
-
----
 
 ## Investigated but not included
 
@@ -647,6 +606,15 @@ smaller SSVTP contribution (~10K@240×320).
   [`yxma/gelsight-mini-pretrain-nc`](https://huggingface.co/datasets/yxma/gelsight-mini-pretrain-nc),
   where it is now superseded by `sparsh` — the same data at a larger
   upstream snapshot.)
+- **TVL — Touch-Vision-Language** (Fu et al., ICML 2024,
+  [project](https://tactile-vlm.github.io/),
+  [dataset](https://huggingface.co/datasets/mlfu7/Touch-Vision-Language-Dataset))
+  — **uses a DIGIT sensor, not GelSight Mini.** Quote from project page:
+  *"Tactile data are collected using a DIGIT sensor: a compact, open-
+  source tactile sensor that provides observations in the form of RGB
+  images."* We initially ingested 209,795 frames before noticing the
+  sensor mismatch. Parquet preserved locally for potential future
+  `digit-pretrain` repo.
 - **Touch-and-Go** (Yang et al., NeurIPS 2022, [GitHub](https://github.com/fredfyyang/Touch-and-Go)) —
   **uses a different GelSight model, not the Mini.** The dataset's
   `gelsight.mp4` clips show 640×480 frames with white-light illumination

unit/train-00000-of-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:05f06fa8aa20bdf9f3c79d0866a830e6a062d89831c5ed25101df3596d9f06ad
-size 7870341
+oid sha256:240b5a0613cf00681ddf235a678191c77b0315f84564348ee59d8be100efc635
+size 6536744