andandandand commited on
Commit
991fbda
·
verified ·
1 Parent(s): d238266

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +120 -76
README.md CHANGED
@@ -12,12 +12,11 @@ tags:
12
  - fiftyone
13
  - image
14
  - image-classification
15
- dataset_summary: '
16
-
17
-
18
-
19
-
20
- This is a [FiftyOne](https://github.com/voxel51/fiftyone) dataset with 169 samples.
21
 
22
 
23
  ## Installation
@@ -63,11 +62,7 @@ dataset_summary: '
63
 
64
  <!-- Provide a quick summary of the dataset. -->
65
 
66
-
67
-
68
-
69
-
70
- This is a [FiftyOne](https://github.com/voxel51/fiftyone) dataset with 169 samples.
71
 
72
  ## Installation
73
 
@@ -91,135 +86,184 @@ dataset = load_from_hub("andandandand/bo_or_not")
91
  session = fo.launch_app(dataset)
92
  ```
93
 
94
-
95
  ## Dataset Details
96
 
97
  ### Dataset Description
98
 
99
- <!-- Provide a longer summary of what this dataset is. -->
100
-
101
 
 
102
 
103
- - **Curated by:** [More Information Needed]
104
- - **Funded by [optional]:** [More Information Needed]
105
- - **Shared by [optional]:** [More Information Needed]
106
- - **Language(s) (NLP):** en
107
  - **License:** mit
 
 
 
108
 
109
- ### Dataset Sources [optional]
110
 
111
- <!-- Provide the basic links for the dataset. -->
112
-
113
- - **Repository:** [More Information Needed]
114
- - **Paper [optional]:** [More Information Needed]
115
- - **Demo [optional]:** [More Information Needed]
116
 
117
  ## Uses
118
 
119
- <!-- Address questions around how the dataset is intended to be used. -->
120
-
121
  ### Direct Use
122
 
123
- <!-- This section describes suitable use cases for the dataset. -->
124
-
125
- [More Information Needed]
 
 
126
 
127
  ### Out-of-Scope Use
128
 
129
- <!-- This section addresses misuse, malicious use, and uses that the dataset will not work well for. -->
130
-
131
- [More Information Needed]
 
 
132
 
133
  ## Dataset Structure
134
 
135
- <!-- This section provides a description of the dataset fields, and additional information about the dataset structure such as criteria used to create the splits, relationships between data points, etc. -->
 
 
 
 
 
 
 
 
 
 
 
 
 
136
 
137
- [More Information Needed]
 
 
 
 
 
138
 
139
  ## Dataset Creation
140
 
141
  ### Curation Rationale
142
 
143
- <!-- Motivation for the creation of this dataset. -->
144
-
145
- [More Information Needed]
 
 
 
146
 
147
  ### Source Data
148
 
149
- <!-- This section describes the source data (e.g. news text and headlines, social media posts, translated sentences, ...). -->
150
-
151
  #### Data Collection and Processing
152
 
153
- <!-- This section describes the data collection and processing process such as data selection criteria, filtering and normalization methods, tools and libraries used, etc. -->
 
 
154
 
155
- [More Information Needed]
 
 
 
 
156
 
157
  #### Who are the source data producers?
158
 
159
- <!-- This section describes the people or systems who originally created the data. It should also include self-reported demographic or identity information for the source data creators if this information is available. -->
160
-
161
- [More Information Needed]
162
-
163
- ### Annotations [optional]
164
 
165
- <!-- If the dataset contains annotations which are not part of the initial data collection, use this section to describe them. -->
166
 
167
  #### Annotation process
168
 
169
- <!-- This section describes the annotation process such as annotation tools used in the process, the amount of data annotated, annotation guidelines provided to the annotators, interannotator statistics, annotation validation, etc. -->
 
 
170
 
171
- [More Information Needed]
172
 
173
  #### Who are the annotators?
174
 
175
- <!-- This section describes the people or systems who created the annotations. -->
176
 
177
- [More Information Needed]
178
 
179
- #### Personal and Sensitive Information
180
 
181
- <!-- State whether the dataset contains data that might be considered personal, sensitive, or private (e.g., data that reveals addresses, uniquely identifiable names or aliases, racial or ethnic origins, sexual orientations, religious beliefs, political opinions, financial or health data, etc.). If efforts were made to anonymize the data, describe the anonymization process. -->
 
 
 
 
 
 
182
 
183
- [More Information Needed]
184
 
185
- ## Bias, Risks, and Limitations
186
-
187
- <!-- This section is meant to convey both technical and sociotechnical limitations. -->
 
 
188
 
189
- [More Information Needed]
190
-
191
- ### Recommendations
192
 
193
- <!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
194
 
195
- Users should be made aware of the risks, biases and limitations of the dataset. More information needed for further recommendations.
 
 
 
196
 
197
- ## Citation [optional]
198
 
199
- <!-- If there is a paper or blog post introducing the dataset, the APA and Bibtex information for that should go in this section. -->
 
 
200
 
201
- **BibTeX:**
202
 
203
- [More Information Needed]
 
 
 
 
204
 
205
- **APA:**
206
 
207
- [More Information Needed]
208
 
209
- ## Glossary [optional]
 
 
 
 
210
 
211
- <!-- If relevant, include terms and calculations in this section that can help readers understand the dataset or dataset card. -->
 
 
 
212
 
213
- [More Information Needed]
214
 
215
- ## More Information [optional]
 
 
 
 
216
 
217
- [More Information Needed]
218
 
219
- ## Dataset Card Authors [optional]
220
 
221
- [More Information Needed]
222
 
223
  ## Dataset Card Contact
224
 
225
- [More Information Needed]
 
12
  - fiftyone
13
  - image
14
  - image-classification
15
+ - transfer-learning
16
+ - vgg16
17
+ - binary-classification
18
+ - portuguese-water-dog
19
+ dataset_summary: 'This is a [FiftyOne](https://github.com/voxel51/fiftyone) dataset with 169 samples for binary classification of Bo (Barack Obama''s Portuguese Water Dog) vs other pets. The dataset demonstrates transfer learning using a pre-trained VGG16 model adapted for this specific classification task.
 
20
 
21
 
22
  ## Installation
 
62
 
63
  <!-- Provide a quick summary of the dataset. -->
64
 
65
+ This is a [FiftyOne](https://github.com/voxel51/fiftyone) dataset with 169 samples for binary classification of Bo (Barack Obama's Portuguese Water Dog) vs other pets. The dataset demonstrates transfer learning using a pre-trained VGG16 model adapted for this specific classification task.
 
 
 
 
66
 
67
  ## Installation
68
 
 
86
  session = fo.launch_app(dataset)
87
  ```
88
 
 
89
  ## Dataset Details
90
 
91
  ### Dataset Description
92
 
93
+ This dataset contains images for a binary classification task designed to distinguish Bo, Barack Obama's Portuguese Water Dog, from other pets (cats and dogs). The dataset was created to explore transfer learning techniques using a pre-trained VGG16 model. Bo was the Obama family's pet during their time in the White House, and this dataset simulates a computer vision system that the Secret Service might have used to automatically recognize him.
 
94
 
95
+ The dataset includes both original images and augmented versions created through various transformations including affine transforms, elastic deformations, perspective changes, and color adjustments to improve model generalization.
96
 
97
+ - **Curated by:** Educational project demonstrating transfer learning and computer vision techniques
98
+ - **Language:** en
 
 
99
  - **License:** mit
100
+ - **Task:** Binary image classification (Bo vs Not Bo)
101
+ - **Model Architecture:** VGG16 with modified classifier layers
102
+ - **Total Samples:** 169 images
103
 
104
+ ### Dataset Sources
105
 
106
+ - **Repository:** Part of practical computer vision educational materials
107
+ - **Base Model:** VGG16 pre-trained on ImageNet
108
+ - **Original Data:** Google Drive collection of Bo images and other pet images
 
 
109
 
110
  ## Uses
111
 
 
 
112
  ### Direct Use
113
 
114
+ This dataset is primarily intended for:
115
+ - **Educational purposes**: Learning transfer learning concepts and binary classification
116
+ - **Computer vision research**: Exploring fine-tuning techniques with pre-trained models
117
+ - **FiftyOne demonstrations**: Showcasing dataset management, visualization, and model evaluation capabilities
118
+ - **Binary classification experiments**: Testing different architectures and hyperparameters
119
 
120
  ### Out-of-Scope Use
121
 
122
+ This dataset should not be used for:
123
+ - **Production pet identification systems**: The dataset is too small and specific for robust real-world applications
124
+ - **General dog breed classification**: Limited to distinguishing one specific dog from other pets
125
+ - **Commercial applications**: Dataset size and scope are insufficient for commercial deployment
126
+ - **Sensitive security applications**: This is a demonstration dataset, not suitable for actual security systems
127
 
128
  ## Dataset Structure
129
 
130
+ The dataset is organized with the following structure:
131
+
132
+ - **Classes**: 2 classes
133
+ - `bo`: Images of Bo (Barack Obama's Portuguese Water Dog)
134
+ - `not_bo`: Images of other pets (cats and dogs)
135
+ - **Splits**:
136
+ - Training: ~50% of training/validation data
137
+ - Validation: ~50% of training/validation data
138
+ - Test: Independent test set
139
+ - **Image Format**: RGB images
140
+ - **Preprocessing**: Images are resized to 224x224 pixels with ImageNet normalization
141
+ - **Augmentations**: Training data includes affine transforms, elastic deformations, perspective changes, brightness/contrast adjustments, and horizontal flips
142
+
143
+ ### Sample Distribution
144
 
145
+ The dataset contains samples tagged with:
146
+ - `train`: Training samples
147
+ - `validation`: Validation samples
148
+ - `test`: Test samples
149
+ - `original`: Original unaugmented images
150
+ - `augmented`: Augmented training samples
151
 
152
  ## Dataset Creation
153
 
154
  ### Curation Rationale
155
 
156
+ This dataset was created as an educational tool to demonstrate:
157
+ 1. **Transfer Learning**: How to adapt a pre-trained model (VGG16) for a new classification task
158
+ 2. **Binary Classification**: Implementing a specific classification problem with practical relevance
159
+ 3. **Data Augmentation**: Techniques to improve model generalization with limited data
160
+ 4. **FiftyOne Integration**: Dataset management, visualization, and evaluation workflows
161
+ 5. **Model Evaluation**: Comprehensive assessment including confusion matrices and classification reports
162
 
163
  ### Source Data
164
 
 
 
165
  #### Data Collection and Processing
166
 
167
+ The source data consists of:
168
+ - **Bo Images**: Photographs of Bo, Barack Obama's Portuguese Water Dog, a Portuguese Water Dog breed
169
+ - **Other Pet Images**: Collection of cats and dogs labeled as `not_bo`
170
 
171
+ Processing steps included:
172
+ 1. **Image standardization**: All images converted to RGB format
173
+ 2. **Augmentation**: Training images enhanced with geometric and photometric transformations
174
+ 3. **Normalization**: ImageNet mean and standard deviation applied for VGG16 compatibility
175
+ 4. **Quality control**: Manual verification of labels and image quality
176
 
177
  #### Who are the source data producers?
178
 
179
+ The source data appears to be curated from publicly available images, with Bo's images likely sourced from official White House photography and news media coverage during the Obama administration.
 
 
 
 
180
 
181
+ ### Annotations
182
 
183
  #### Annotation process
184
 
185
+ Images were manually labeled into two categories:
186
+ - `bo`: Images containing Bo, the Portuguese Water Dog
187
+ - `not_bo`: Images containing other pets (cats and various dog breeds)
188
 
189
+ The annotation process focused on creating a balanced binary classification task suitable for transfer learning demonstrations.
190
 
191
  #### Who are the annotators?
192
 
193
+ Annotations were created as part of the educational project development, with manual labeling to ensure accuracy for the binary classification task.
194
 
195
+ ## Technical Implementation
196
 
197
+ ### Model Architecture
198
 
199
+ - **Base Model**: VGG16 pre-trained on ImageNet
200
+ - **Modifications**:
201
+ - Frozen convolutional layers (feature extraction)
202
+ - Custom classifier: Linear(25088 → 4096) → ReLU → Linear(4096 → 1)
203
+ - **Loss Function**: Binary Cross Entropy with Logits
204
+ - **Optimizer**: Adam (lr=0.003, betas=(0.9, 0.999))
205
+ - **Training Epochs**: 10 epochs with early stopping
206
 
207
+ ### Performance Metrics
208
 
209
+ The model evaluation includes:
210
+ - **Loss tracking**: Training and validation loss curves
211
+ - **Classification report**: Precision, recall, F1-score for both classes
212
+ - **Confusion matrix**: Visual representation of classification performance
213
+ - **Confidence scores**: Probability estimates for each prediction
214
 
215
+ ## Bias, Risks, and Limitations
 
 
216
 
217
+ ### Technical Limitations
218
 
219
+ - **Small dataset size**: 169 samples may not capture full variability
220
+ - **Limited breed diversity**: Focus on one specific dog vs. general "other pets"
221
+ - **Domain specificity**: Trained specifically for Bo recognition, not generalizable to other Portuguese Water Dogs
222
+ - **Image quality dependency**: Performance may vary with different lighting, angles, or image quality
223
 
224
+ ### Potential Biases
225
 
226
+ - **Breed bias**: Model may learn Portuguese Water Dog features rather than Bo-specific characteristics
227
+ - **Environmental bias**: Training images may reflect specific environments or photography conditions
228
+ - **Size bias**: Limited sample size may not represent full range of poses, ages, or conditions
229
 
230
+ ### Recommendations
231
 
232
+ Users should be aware that:
233
+ 1. **This is a demonstration dataset**: Not suitable for production applications
234
+ 2. **Limited generalization**: Results may not transfer to other similar classification tasks
235
+ 3. **Educational focus**: Primary value is in learning transfer learning concepts
236
+ 4. **Expansion needed**: Real-world applications would require significantly more diverse data
237
 
238
+ ## Citation
239
 
240
+ If you use this dataset in your research or educational projects, please acknowledge its educational purpose and the demonstration of transfer learning techniques with VGG16 and FiftyOne.
241
 
242
+ **Usage Context:**
243
+ ```python
244
+ # Example educational usage
245
+ import fiftyone as fo
246
+ from fiftyone.utils.huggingface import load_from_hub
247
 
248
+ # Load for learning transfer learning concepts
249
+ dataset = load_from_hub("andandandand/bo_or_not")
250
+ session = fo.launch_app(dataset)
251
+ ```
252
 
253
+ ## More Information
254
 
255
+ This dataset is part of practical computer vision educational materials demonstrating:
256
+ - Transfer learning with PyTorch and torchvision
257
+ - FiftyOne dataset management and visualization
258
+ - Binary classification model development and evaluation
259
+ - Data augmentation techniques for improving model performance
260
 
261
+ For more details on the implementation, see the accompanying Jupyter notebook which provides step-by-step explanations of the transfer learning process, model architecture modifications, and evaluation techniques.
262
 
263
+ ## Dataset Card Authors
264
 
265
+ Created as part of practical computer vision educational materials focusing on transfer learning and binary classification techniques.
266
 
267
  ## Dataset Card Contact
268
 
269
+ For questions about this educational dataset, refer to the practical computer vision course materials or the accompanying implementation notebook.