snnn001 commited on
Commit
d321bda
·
verified ·
1 Parent(s): c2eeb80

Add LiteRT converted swin_small_patch4_window7_224

Browse files
Files changed (2) hide show
  1. README.md +60 -0
  2. model.tflite +3 -0
README.md ADDED
@@ -0,0 +1,60 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ library_name: litert
3
+ base_model: timm/swin_small_patch4_window7_224.ms_in22k_ft_in1k
4
+ tags:
5
+ - vision
6
+ - image-classification
7
+ datasets:
8
+ - imagenet-1k
9
+ ---
10
+ # swin_small_patch4_window7_224
11
+
12
+ Converted TIMM image classification model for LiteRT.
13
+
14
+ - Source architecture: `swin_small_patch4_window7_224`
15
+ - Source checkpoint: `timm/swin_small_patch4_window7_224.ms_in22k_ft_in1k`
16
+ - File: `model.tflite`
17
+ - Input: `float32` tensor in NCHW layout, shape `[1, 3, 224, 224]`
18
+ - Output: ImageNet-1K logits, shape `[1, 1000]`
19
+
20
+ ## Runtime Status
21
+
22
+ - CPU smoke test: passed with LiteRT `CompiledModel`.
23
+ - GPU delegation: currently blocked for this model by rank-5 tensor patterns in the GPU backend, mostly `RESHAPE`, `TRANSPOSE`, and related window/attention operations. The model is published as CPU-ready while GPU support is being improved.
24
+
25
+ ## Model Details
26
+
27
+ - **Model Type:** Image classification / feature backbone
28
+ - **Model Stats:**
29
+ - Params (M): 49.6
30
+ - GMACs: 8.8
31
+ - Activations (M): 27.5
32
+ - Image size: 224 x 224
33
+ - **Papers:**
34
+ - Swin Transformer: Hierarchical Vision Transformer using Shifted Windows: https://arxiv.org/abs/2103.14030
35
+ - **Original:** https://github.com/microsoft/Swin-Transformer
36
+ - **Dataset:** ImageNet-1k
37
+ - **Pretrain Dataset:** ImageNet-22k
38
+
39
+ ## Citation
40
+
41
+ ```bibtex
42
+ @inproceedings{liu2021Swin,
43
+ title={Swin Transformer: Hierarchical Vision Transformer using Shifted Windows},
44
+ author={Liu, Ze and Lin, Yutong and Cao, Yue and Hu, Han and Wei, Yixuan and Zhang, Zheng and Lin, Stephen and Guo, Baining},
45
+ booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
46
+ year={2021}
47
+ }
48
+ ```
49
+ ```bibtex
50
+ @misc{rw2019timm,
51
+ author = {Ross Wightman},
52
+ title = {PyTorch Image Models},
53
+ year = {2019},
54
+ publisher = {GitHub},
55
+ journal = {GitHub repository},
56
+ doi = {10.5281/zenodo.4414861},
57
+ howpublished = {\url{https://github.com/huggingface/pytorch-image-models}}
58
+ }
59
+ ```
60
+
model.tflite ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8d17ce2a11ceb2610a3966ca2de940e3090f8f373625a1a8c7837ef26e1240d5
3
+ size 202264672