image_bytes unknown | action stringclasses 325
values | game stringclasses 2
values | trial_id int32 160 551 | frame_idx int32 0 0 | image_size int32 384 384 |
|---|---|---|---|---|---|
[
137,
80,
78,
71,
13,
10,
26,
10,
0,
0,
0,
13,
73,
72,
68,
82,
0,
0,
1,
128,
0,
0,
1,
128,
8,
2,
0,
0,
0,
43,
165,
34,
232,
0,
0,
127,
101,
73,
68,
65,
84,
120,
218,
236,
253,
249,
147,
229,
89,
118,
31,
134,
125,... | <|action_start|> LEFT ; LEFT ; LEFT <|action_end|> | alien | 323 | 0 | 384 |
[
137,
80,
78,
71,
13,
10,
26,
10,
0,
0,
0,
13,
73,
72,
68,
82,
0,
0,
1,
128,
0,
0,
1,
128,
8,
2,
0,
0,
0,
43,
165,
34,
232,
0,
0,
126,
197,
73,
68,
65,
84,
120,
218,
236,
253,
233,
147,
101,
75,
114,
31,
136,
253,... | <|action_start|> LEFT ; LEFT ; LEFT <|action_end|> | alien | 323 | 0 | 384 |
[
137,
80,
78,
71,
13,
10,
26,
10,
0,
0,
0,
13,
73,
72,
68,
82,
0,
0,
1,
128,
0,
0,
1,
128,
8,
2,
0,
0,
0,
43,
165,
34,
232,
0,
0,
124,
220,
73,
68,
65,
84,
120,
218,
237,
253,
233,
147,
108,
73,
118,
31,
136,
253,... | <|action_start|> LEFT ; LEFT ; LEFT <|action_end|> | alien | 323 | 0 | 384 |
[
137,
80,
78,
71,
13,
10,
26,
10,
0,
0,
0,
13,
73,
72,
68,
82,
0,
0,
1,
128,
0,
0,
1,
128,
8,
2,
0,
0,
0,
43,
165,
34,
232,
0,
0,
123,
201,
73,
68,
65,
84,
120,
218,
237,
253,
233,
183,
101,
73,
118,
31,
134,
253,... | <|action_start|> LEFT ; LEFT ; LEFT <|action_end|> | alien | 323 | 0 | 384 |
"iVBORw0KGgoAAAANSUhEUgAAAYAAAAGACAIAAAArpSLoAAB83ElEQVR42u396ZNsSXYfiP3Ocfd7b0Tk/uot9WqvXlC9o7uBJoU(...TRUNCATED) | <|action_start|> LEFT ; LEFT ; LEFT <|action_end|> | alien | 323 | 0 | 384 |
"iVBORw0KGgoAAAANSUhEUgAAAYAAAAGACAIAAAArpSLoAACAd0lEQVR42uz96ZdlS3Yfhv32johzzp1yrPHNc0/oxtgAJxAEJ9O(...TRUNCATED) | <|action_start|> LEFT ; LEFT ; LEFT <|action_end|> | alien | 323 | 0 | 384 |
"iVBORw0KGgoAAAANSUhEUgAAAYAAAAGACAIAAAArpSLoAAB/ZUlEQVR42uz9+ZPlWXYfhn3Oufe7vCXX2qv3nu5Bzw7MAAMKAgh(...TRUNCATED) | <|action_start|> LEFT ; LEFT ; LEFT <|action_end|> | alien | 323 | 0 | 384 |
"iVBORw0KGgoAAAANSUhEUgAAAYAAAAGACAIAAAArpSLoAAB+xUlEQVR42uz96ZNlS3IfiP3cI+Kcc7dca3v73isaja0BEgQIgps(...TRUNCATED) | <|action_start|> LEFT ; LEFT ; LEFT <|action_end|> | alien | 323 | 0 | 384 |
"iVBORw0KGgoAAAANSUhEUgAAAYAAAAGACAIAAAArpSLoAAB67klEQVR42u396ZNlSXYfiP3Ocfd773sv9qysrMraqxdWd6O70Q2(...TRUNCATED) | <|action_start|> LEFT ; LEFT ; LEFT <|action_end|> | alien | 323 | 0 | 384 |
"iVBORw0KGgoAAAANSUhEUgAAAYAAAAGACAIAAAArpSLoAAB9w0lEQVR42u396ZdlS3Yfhv32johzzh1yrunNQ7/XIxoNkN0ABRI(...TRUNCATED) | <|action_start|> LEFT ; LEFT ; LEFT <|action_end|> | alien | 323 | 0 | 384 |
End of preview. Expand in Data Studio
TESS-Atari Stage 1 - Preprocessed (15Hz, 384x384)
Training-ready version of the 15Hz dataset with images pre-resized to 384x384 (SmolVLM native resolution).
Overview
| Metric | Value |
|---|---|
| Source | TESS-Computer/atari-vla-stage1-15hz |
| Samples | 1,340,293 |
| Image Size | 384x384 (pre-resized) |
| Action Rate | 15 Hz (3 actions per observation) |
| Format | Lumine-style action tokens |
Why Preprocessed?
Training VLMs requires resizing images to the model's native resolution. Doing this on-the-fly creates a CPU bottleneck. This dataset has images already resized, giving ~10x faster training:
Raw dataset: 160x210 → resize during training → slow (CPU bound)
Preprocessed: 384x384 → ready to use → fast (GPU saturated)
Action Format
<|action_start|> RIGHT ; RIGHT ; FIRE <|action_end|>
<|action_start|> LEFT ; LEFT ; LEFT <|action_end|>
<|action_start|> NOOP ; UP ; UPFIRE <|action_end|>
Schema
| Field | Type | Description |
|---|---|---|
image_bytes |
bytes | PNG at 384x384 (pre-resized) |
action |
string | Lumine-style chunked action token |
game |
string | Game name |
trial_id |
int | Human player trial number |
frame_idx |
int | Frame index in trial |
image_size |
int | Always 384 |
Usage
from datasets import load_dataset
from PIL import Image
from io import BytesIO
# Load preprocessed dataset
ds = load_dataset("TESS-Computer/tess-atari-15hz-384", split="train")
# Images are already 384x384 - no resizing needed!
sample = ds[0]
img = Image.open(BytesIO(sample["image_bytes"]))
print(img.size) # (384, 384)
print(sample["action"]) # <|action_start|> LEFT ; LEFT ; LEFT <|action_end|>
Training
python scripts/train_v2.py \
--preprocessed TESS-Computer/tess-atari-15hz-384 \
--epochs 3 \
--batch-size 4 \
--grad-accum 32 \
--wandb \
--push-to-hub
Related
- Raw 15Hz dataset - Original with 160x210 images
- Raw 5Hz dataset - Single action per observation
- TESS-Atari repo - Training code
Citation
@misc{tessatari2025,
title={TESS-Atari: Vision-Language-Action Models for Atari Games},
author={Lezzaik, Hussein},
year={2025},
url={https://github.com/HusseinLezzaik/TESS-Atari}
}
@misc{atarihead2019,
title={Atari-HEAD: Atari Human Eye-Tracking and Demonstration Dataset},
author={Zhang, Ruohan and others},
year={2019},
url={https://zenodo.org/records/3451402}
}
- Downloads last month
- 22