Hugging Face's logo

Datasets:
issai
/
TFW

Tasks:
Object Detection
Modalities:
Image
Size:
1K<n<10K
Tags:
thermal faces
thermal facial landmarks
License:
Dataset card Data Studio Files Community
rassulya commited on
Commit
615ecae
·
verified ·
1 Parent(s): 7111d8b

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +34 -34
README.md CHANGED
@@ -2,24 +2,46 @@
2
 
3
  **Repository:** [https://github.com/IS2AI/TFW](https://github.com/IS2AI/TFW)
4
 
5
- **Summary Description:**
6
 
7
- The TFW dataset comprises thermal images captured in controlled indoor, semi-controlled indoor, and uncontrolled outdoor environments. It's a multi-environment dataset, leveraging a previously published SpeakingFaces dataset for its controlled indoor component. The remaining images were acquired using a FLIR T540 thermal camera. Each image is manually annotated with bounding boxes for faces and five facial landmarks. The dataset is valuable for thermal face recognition research and applications.
8
 
 
9
 
10
- **Summary of Abstract (Unavailable):**
 
 
 
 
 
11
 
12
- The abstract from the linked TechRxiv preprint was inaccessible.
13
 
 
14
 
15
- **Dataset Statistics:**
16
 
17
- | Environment | Subjects | Images | Labeled Faces | Visual Pair |
18
- |---|---|---|---|---|
19
- | c-indoor | 142 | 5,112 | 5,112 | yes |
20
- | s-indoor | 9 | 780 | 1,748 | yes |
21
- | u-outdoor | 15 | 4,090 | 9,649 | no |
22
- | combined | 147 | 9,982 | 16,509 | yes & no |
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
23
 
24
 
25
  **Citation:**
@@ -36,26 +58,4 @@ The abstract from the linked TechRxiv preprint was inaccessible.
36
  doi={10.1109/TIFS.2022.3177949}}
37
  ```
38
 
39
- **Pre-trained Models Table:**
40
-
41
- | Model | Backbone | c-indoor AP<sub>50</sub> | u-outdoor AP<sub>50</sub> | Speed (ms) V100 b1 | Params (M) | Flops (G) @512x384 |
42
- |---|---|---|---|---|---|---|
43
- | YOLOv5n | CSPNet | 100 | 97.29 | 6.16 | 1.76 | 0.99 |
44
- | YOLOv5n6 | CSPNet | 100 | 95.79 | 8.18 | 3.09 | 1.02 |
45
- | YOLOv5s | CSPNet | 100 | 96.82 | 7.20 | 7.05 | 3.91 |
46
- | YOLOv5s6 | CSPNet | 100 | 96.83 | 9.05 | 12.31 | 3.88 |
47
- | YOLOv5m | CSPNet | 100 | 97.16 | 9.59 | 21.04 | 12.07 |
48
- | YOLOv5m6 | CSPNet | 100 | 97.10 | 12.11 | 35.25 | 11.76 |
49
- | YOLOv5l | CSPNet | 100 | 96.68 | 12.39 | 46.60 | 27.38 |
50
- | YOLOv5l6 | CSPNet | 100 | 96.29 | 15.73 | 76.16 | 110.2 |
51
- | YOLOv5n-Face | ShuffleNetv2 | 100 | 95.93 | 10.12 | 1.72 | 1.36 |
52
- | YOLOv5n6-Face | ShuffleNetv2 | 100 | 95.59 | 13.30 | 2.54 | 1.38 |
53
- | YOLOv5s-Face | CSPNet | 100 | 96.73 | 8.29 | 7.06 | 3.67 |
54
- | YOLOv5s6-Face | CSPNet | 100 | 96.36 | 10.86 | 12.37 | 3.75 |
55
- | YOLOv5m-Face | CSPNet | 100 | 95.32 | 11.01 | 21.04 | 11.58 |
56
- | YOLOv5m6-Face | CSPNet | 100 | 96.32 | 13.97 | 35.45 | 11.84 |
57
- | YOLOv5l-Face | CSPNet | 100 | 96.18 | 13.57 | 46.59 | 25.59 |
58
- | YOLOv5l6-Face | CSPNet | 100 | 95.76 | 17.29 | 76.67 | 113.2 |
59
-
60
-
61
- **(Note: Links to pre-trained models and example images are omitted as requested.)**
 
2
 
3
  **Repository:** [https://github.com/IS2AI/TFW](https://github.com/IS2AI/TFW)
4
 
5
+ **Summary:**
6
 
7
+ The TFW dataset provides a collection of thermal face images captured in diverse environments: controlled indoor (c-indoor), semi-controlled indoor (s-indoor), and uncontrolled outdoor (u-outdoor). It's annotated with bounding boxes and 5-point facial landmarks. The dataset includes images from the SpeakingFaces dataset ([https://github.com/IS2AI/SpeakingFaces](https://github.com/IS2AI/SpeakingFaces)) for the c-indoor subset, and additional data collected using a FLIR T540 thermal camera for s-indoor and u-outdoor subsets. The paper detailing the dataset and accompanying pre-trained YOLOv5 and YOLOv5Face models are also available.
8
 
9
+ **Dataset Statistics:**
10
 
11
+ | Environment | Subjects | Images | Labeled Faces | Visual Pair |
12
+ |--------------|---------|--------|-----------------|-------------|
13
+ | c-indoor | 142 | 5,112 | 5,112 | yes |
14
+ | s-indoor | 9 | 780 | 1,748 | yes |
15
+ | u-outdoor | 15 | 4,090 | 9,649 | no |
16
+ | **combined** | **147** | **9,982** | **16,509** | yes & no |
17
 
 
18
 
19
+ **Example Images:**
20
 
21
+ [Image would be displayed here. Source: https://github.com/IS2AI/TFW/blob/main/figures/example.png]
22
 
23
+ **Pre-trained Models:**
24
+
25
+ The following table summarizes the performance of pre-trained YOLOv5 and YOLOv5Face models on the TFW dataset. Note that links to the models are not included here as per the instructions.
26
+
27
+ | Model | Backbone | c-indoor AP<sub>50</sub> | u-outdoor AP<sub>50</sub> | Speed (ms) V100 b1 | Params (M) | Flops (G) @512x384 |
28
+ |-----------------|--------------|------------------------|------------------------|----------------------|-------------|--------------------|
29
+ | YOLOv5n | CSPNet | 100 | 97.29 | 6.16 | 1.76 | 0.99 |
30
+ | YOLOv5n6 | CSPNet | 100 | 95.79 | 8.18 | 3.09 | 1.02 |
31
+ | YOLOv5s | CSPNet | 100 | 96.82 | 7.20 | 7.05 | 3.91 |
32
+ | YOLOv5s6 | CSPNet | 100 | 96.83 | 9.05 | 12.31 | 3.88 |
33
+ | YOLOv5m | CSPNet | 100 | 97.16 | 9.59 | 21.04 | 12.07 |
34
+ | YOLOv5m6 | CSPNet | 100 | 97.10 | 12.11 | 35.25 | 11.76 |
35
+ | YOLOv5l | CSPNet | 100 | 96.68 | 12.39 | 46.60 | 27.38 |
36
+ | YOLOv5l6 | CSPNet | 100 | 96.29 | 15.73 | 76.16 | 110.2 |
37
+ | YOLOv5n-Face | ShuffleNetv2 | 100 | 95.93 | 10.12 | 1.72 | 1.36 |
38
+ | YOLOv5n6-Face | ShuffleNetv2 | 100 | 95.59 | 13.30 | 2.54 | 1.38 |
39
+ | YOLOv5s-Face | CSPNet | 100 | 96.73 | 8.29 | 7.06 | 3.67 |
40
+ | YOLOv5s6-Face | CSPNet | 100 | 96.36 | 10.86 | 12.37 | 3.75 |
41
+ | YOLOv5m-Face | CSPNet | 100 | 95.32 | 11.01 | 21.04 | 11.58 |
42
+ | YOLOv5m6-Face | CSPNet | 100 | 96.32 | 13.97 | 35.45 | 11.84 |
43
+ | YOLOv5l-Face | CSPNet | 100 | 96.18 | 13.57 | 46.59 | 25.59 |
44
+ | YOLOv5l6-Face | CSPNet | 100 | 95.76 | 17.29 | 76.67 | 113.2 |
45
 
46
 
47
  **Citation:**
 
58
  doi={10.1109/TIFS.2022.3177949}}
59
  ```
60
 
61
+ **(Note: Example images and demonstration GIF would be included in the actual Hugging Face dataset card. Links to pre-trained models have been omitted as requested.)**