Update README.md
Browse files
README.md
CHANGED
@@ -59,15 +59,42 @@ the following use cases are recommended for StreetCLIP.
|
|
59 |
## Direct Use
|
60 |
|
61 |
StreetCLIP can be used out-of-the box using zero-shot learning to infer the geolocation of images on a country, region,
|
62 |
-
or city level. Given that StreetCLIP was pretrained on a dataset of
|
63 |
the best performance can be expected on images from a similar distribution.
|
64 |
|
65 |
-
Broader direct use cases
|
|
|
66 |
|
67 |
## Downstream Use
|
68 |
|
69 |
StreetCLIP can be finetuned for any downstream applications that require geographic or street-level urban or rural
|
70 |
-
scene understanding.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
71 |
|
72 |
## Out-of-Scope Use
|
73 |
|
@@ -80,9 +107,8 @@ attempting to geolocalize users' private images
|
|
80 |
|
81 |
## Recommendations
|
82 |
We encourage the community to apply StreetCLIP to applications with significant social impact of which there are many.
|
83 |
-
|
84 |
-
|
85 |
-
environment (i.e. image segmentation, vegetation mapping and classification, tracking deforestation).
|
86 |
|
87 |
## How to Get Started with the Model
|
88 |
|
|
|
59 |
## Direct Use
|
60 |
|
61 |
StreetCLIP can be used out-of-the box using zero-shot learning to infer the geolocation of images on a country, region,
|
62 |
+
or city level. Given that StreetCLIP was pretrained on a dataset of street-level urban and rural images,
|
63 |
the best performance can be expected on images from a similar distribution.
|
64 |
|
65 |
+
Broader direct use cases are any zero-shot image classification tasks that rely on urban and rural street-level
|
66 |
+
understanding or geographical information relating visual clues to their region of origin.
|
67 |
|
68 |
## Downstream Use
|
69 |
|
70 |
StreetCLIP can be finetuned for any downstream applications that require geographic or street-level urban or rural
|
71 |
+
scene understanding. Examples of use cases are the following:
|
72 |
+
|
73 |
+
**Understanding the Built Environment**
|
74 |
+
|
75 |
+
- Analyzing building quality
|
76 |
+
- Building type classifcation
|
77 |
+
- Building energy efficiency Classification
|
78 |
+
|
79 |
+
**Analyzing Infrastructure**
|
80 |
+
|
81 |
+
- Analyzing road quality
|
82 |
+
- Utility pole maintenance
|
83 |
+
- Identifying damage from natural disasters or armed conflicts
|
84 |
+
|
85 |
+
**Understanding the Natural Environment**
|
86 |
+
|
87 |
+
- Mapping vegetation
|
88 |
+
- Vegetation classification
|
89 |
+
- Soil type classifcation
|
90 |
+
- Tracking deforestation
|
91 |
+
|
92 |
+
**General Use Cases**
|
93 |
+
|
94 |
+
- Street-level image segmentation
|
95 |
+
- Urban and rural scene classification
|
96 |
+
- Object detection in urban or rural environments
|
97 |
+
- Improving navigation and self-driving car technology
|
98 |
|
99 |
## Out-of-Scope Use
|
100 |
|
|
|
107 |
|
108 |
## Recommendations
|
109 |
We encourage the community to apply StreetCLIP to applications with significant social impact of which there are many.
|
110 |
+
The first three categories of potential use cases under Downstream Use list potential use cases with social impact
|
111 |
+
to explore.
|
|
|
112 |
|
113 |
## How to Get Started with the Model
|
114 |
|