carohiguera commited on
Commit
dbf4fbc
1 Parent(s): de2e534

added Sparsh IJEPA small

Browse files
Files changed (3) hide show
  1. README.md +47 -0
  2. ijepa_vitsmall.ckpt +3 -0
  3. ijepa_vitsmall.safetensors +3 -0
README.md ADDED
@@ -0,0 +1,47 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: cc-by-nc-4.0
3
+ tags:
4
+ - sparsh
5
+ - ijepa
6
+ - small
7
+ - tactile
8
+ ---
9
+
10
+ # Sparsh (small-sized model) trained using I-JEPA
11
+
12
+ Sparsh is a Vision Transformer (ViT) model trained using the I-JEPA method, specifically adapted for vision-based tactile sensors such as DIGIT and GelSight.
13
+
14
+ Disclaimer: This model card was written by the Sparsh authors. The ViT model and I-JEPA objectives have been adapted for the tactile sensing use case.
15
+
16
+ ## Model description
17
+ We introduce *Sparsh*, a family of touch representations trained using Self-Supervised Learning (SSL) across multiple sensors, including DIGIT, GelSight 2017 (with markers), and GelSight Mini (without markers). This model was trained using the I-JEPA SSL approach.
18
+
19
+ The model takes two tactile images as input, with a temporal stride of 5 samples across the channel dimension, $I_t ⊕ I_{t−5} → x ∈ R^{h×w×6}$. For a sensor operating at 60FPS, this corresponds to an inference window of approximately 80ms, which is the reaction time humans need to adjust grip force when detecting partial slip.
20
+
21
+ We preprocess the tactile images by performing background subtraction, which allows for robustness to distractors such as shadows and light placement variations.
22
+
23
+ By pre-training the model via SSL, Sparsh learns representations for pairs of tactile images that can then be used to extract features useful for downstream tasks. To train a downstream task in a supervised fashion, you can place a standard decoder (or head) on top of the pre-trained Sparsh (encoder) by using attentive pooling followed by a shallow MLP.
24
+
25
+
26
+ ## Intended uses & limitations
27
+ You can utilize the Sparsh model to extract touch representations for vision-based tactile sensors, including DIGIT, GelSight, and GelSight mini. You have two options:
28
+
29
+ 1. Use the frozen Sparsh encoder: This allows you to leverage the pre-trained weights of the Sparsh model without modifying them.
30
+ 2. Fine-tune the Sparsh encoder: You can fine-tune the Sparsh encoder along with the training of your downstream task, allowing the model to adapt to your specific use case.
31
+
32
+ Both options enable you to take advantage of the powerful touch representations learned by the Sparsh model.
33
+
34
+ ## How to Use
35
+ For detailed instructions on how to load the encoder and integrate it into your downstream task, please refer to our [GitHub repository](https://github.com/facebookresearch/sparsh).
36
+
37
+ ### BibTeX entry and citation info
38
+ ```bibtex
39
+ @inproceedings{
40
+ higuera2024sparsh,
41
+ title={Sparsh: Self-supervised touch representations for vision-based tactile sensing},
42
+ author={Carolina Higuera and Akash Sharma and Chaithanya Krishna Bodduluri and Taosha Fan and Mrinal Kalakrishnan and Michael Kaess and Byron Boots and Mike Lambeta and Tingfan Wu and Mustafa Mukadam},
43
+ booktitle={8th Annual Conference on Robot Learning},
44
+ year={2024},
45
+ url={https://openreview.net/forum?id=xYJn2e1uu8}
46
+ }
47
+ ```
ijepa_vitsmall.ckpt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:da97b2480ae1d54ae2ceaf320c814d7a1a71c0ff5a92f92760de684683b65e44
3
+ size 410478726
ijepa_vitsmall.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9f0c4285abfbbb5c5e1e6ca20a79f4f7c9f3d0c1cb846853eda4bbb23164720d
3
+ size 87593104