qubvel-hf HF staff commited on
Commit
ba74566
1 Parent(s): 6a7e585

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +46 -104
README.md CHANGED
@@ -22,17 +22,25 @@ widget:
22
  ---
23
 
24
 
25
- # Model Card for Model ID
26
 
27
- <!-- Provide a quick summary of what the model is/does. -->
28
 
 
 
 
 
 
 
 
 
 
29
 
30
 
31
  ## Model Details
32
 
33
- ### Model Description
34
 
35
- The YOLO series has become the most popular framework for real-time object detection due to its reasonable trade-off between speed and accuracy.
36
  However, we observe that the speed and accuracy of YOLOs are negatively affected by the NMS.
37
  Recently, end-to-end Transformer-based detectors (DETRs) have provided an alternative to eliminating NMS.
38
  Nevertheless, the high computational cost limits their practicality and hinders them from fully exploiting the advantage of excluding NMS.
@@ -47,61 +55,32 @@ We also develop scaled RT-DETRs that outperform the lighter YOLO detectors (S an
47
  Furthermore, RT-DETR-R50 outperforms DINO-R50 by 2.2% AP in accuracy and about 21 times in FPS.
48
  After pre-training with Objects365, RT-DETR-R50 / R101 achieves 55.3% / 56.2% AP. The project page: this [https URL](https://zhao-yian.github.io/RTDETR/).
49
 
50
- ![image/png](https://cdn-uploads.huggingface.co/production/uploads/6579e0eaa9e58aec614e9d97/WULSDLsCVs7RNEs9KB0Lr.png)
51
 
52
- This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
 
53
 
54
  - **Developed by:** Yian Zhao and Sangbum Choi
55
- - **Funded by [optional]:** National Key R&D Program of China (No.2022ZD0118201), Natural Science Foundation of China (No.61972217, 32071459, 62176249, 62006133, 62271465),
56
  and the Shenzhen Medical Research Funds in China (No.
57
  B2302037).
58
- - **Shared by [optional]:** Sangbum Choi
59
- - **Model type:**
60
- - **Language(s) (NLP):**
61
  - **License:** Apache-2.0
62
- - **Finetuned from model [optional]:**
63
 
64
- ### Model Sources [optional]
65
 
66
  <!-- Provide the basic links for the model. -->
67
 
 
68
  - **Repository:** https://github.com/lyuwenyu/RT-DETR
69
- - **Paper [optional]:** https://arxiv.org/abs/2304.08069
70
- - **Demo [optional]:** [More Information Needed]
71
-
72
- ## Uses
73
-
74
- <!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
75
-
76
- ### Direct Use
77
-
78
- <!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
79
-
80
- You can use the raw model for object detection. See the [model hub](https://huggingface.co/models?search=rtdetr) to look for all available RTDETR models.
81
-
82
- ### Downstream Use [optional]
83
-
84
- <!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
85
-
86
- ### Out-of-Scope Use
87
-
88
- <!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
89
-
90
- ## Bias, Risks, and Limitations
91
-
92
- <!-- This section is meant to convey both technical and sociotechnical limitations. -->
93
-
94
- ### Recommendations
95
-
96
- <!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
97
-
98
- Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
99
 
100
  ## How to Get Started with the Model
101
 
102
  Use the code below to get started with the model.
103
 
104
- ```
105
  import torch
106
  import requests
107
 
@@ -148,77 +127,48 @@ The RTDETR model was trained on [COCO 2017 object detection](https://cocodataset
148
 
149
  <!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
150
 
151
- We conduct experiments on
152
- COCO [20] and Objects365 [35], where RT-DETR is trained
153
- on COCO train2017 and validated on COCO val2017
154
- dataset. We report the standard COCO metrics, including
155
- AP (averaged over uniformly sampled IoU thresholds ranging from 0.50-0.95 with a step size of 0.05), AP50, AP75, as
156
- well as AP at different scales: APS, APM, APL.
157
 
158
- #### Preprocessing [optional]
159
 
160
- Images are resized/rescaled such that the shortest side is at 640 pixels.
161
 
162
- #### Training Hyperparameters
163
 
164
  - **Training regime:** <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
165
 
166
  ![image/png](https://cdn-uploads.huggingface.co/production/uploads/6579e0eaa9e58aec614e9d97/E15I9MwZCtwNIms-W8Ra9.png)
167
 
168
- #### Speeds, Sizes, Times [optional]
169
-
170
- <!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
171
 
172
  ## Evaluation
173
 
174
- <!-- This section describes the evaluation protocols and provides the results. -->
175
-
176
- This model achieves an AP (average precision) of 53.1 on COCO 2017 validation. For more details regarding evaluation results, we refer to table 2 of the original paper.
177
 
178
- ### Testing Data, Factors & Metrics
 
 
 
 
 
 
 
 
179
 
180
- #### Testing Data
181
 
182
- <!-- This should link to a Dataset Card if possible. -->
183
-
184
- #### Factors
185
-
186
- <!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
187
-
188
- #### Metrics
189
-
190
- <!-- These are the evaluation metrics being used, ideally with a description of why. -->
191
-
192
- ### Results
193
-
194
- #### Summary
195
-
196
-
197
-
198
- ## Model Examination [optional]
199
-
200
- <!-- Relevant interpretability work for the model goes here -->
201
-
202
- ## Environmental Impact
203
-
204
- <!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
205
-
206
- Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
207
-
208
-
209
- ## Technical Specifications [optional]
210
 
211
  ### Model Architecture and Objective
212
 
213
  ![image/png](https://cdn-uploads.huggingface.co/production/uploads/6579e0eaa9e58aec614e9d97/sdIwTRlHNwPzyBNwHja60.png)
214
 
215
- ### Compute Infrastructure
 
 
 
 
216
 
217
- #### Hardware
218
 
219
- #### Software
220
-
221
- ## Citation [optional]
222
 
223
  <!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
224
 
@@ -235,16 +185,8 @@ Carbon emissions can be estimated using the [Machine Learning Impact calculator]
235
  }
236
  ```
237
 
238
- **APA:**
239
-
240
- ## Glossary [optional]
241
-
242
- <!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
243
-
244
- ## More Information [optional]
245
-
246
- ## Model Card Authors [optional]
247
 
248
- [Sangbum Choi](https://huggingface.co/danelcsb)
 
249
 
250
- ## Model Card Contact
 
22
  ---
23
 
24
 
25
+ # Model Card for RT-DETR
26
 
 
27
 
28
+ ## Table of Contents
29
+
30
+ 1. [Model Details](#model-details)
31
+ 2. [Model Sources](#model-sources)
32
+ 3. [How to Get Started with the Model](#how-to-get-started-with-the-model)
33
+ 4. [Training Details](#training-details)
34
+ 5. [Evaluation](#evaluation)
35
+ 6. [Model Architecture and Objective](#model-architecture-and-objective)
36
+ 7. [Citation](#citation)
37
 
38
 
39
  ## Model Details
40
 
41
+ ![image/png](https://cdn-uploads.huggingface.co/production/uploads/6579e0eaa9e58aec614e9d97/WULSDLsCVs7RNEs9KB0Lr.png)
42
 
43
+ > The YOLO series has become the most popular framework for real-time object detection due to its reasonable trade-off between speed and accuracy.
44
  However, we observe that the speed and accuracy of YOLOs are negatively affected by the NMS.
45
  Recently, end-to-end Transformer-based detectors (DETRs) have provided an alternative to eliminating NMS.
46
  Nevertheless, the high computational cost limits their practicality and hinders them from fully exploiting the advantage of excluding NMS.
 
55
  Furthermore, RT-DETR-R50 outperforms DINO-R50 by 2.2% AP in accuracy and about 21 times in FPS.
56
  After pre-training with Objects365, RT-DETR-R50 / R101 achieves 55.3% / 56.2% AP. The project page: this [https URL](https://zhao-yian.github.io/RTDETR/).
57
 
 
58
 
59
+
60
+ This is the model card of a 🤗 [transformers](https://huggingface.co/docs/transformers/index) model that has been pushed on the Hub.
61
 
62
  - **Developed by:** Yian Zhao and Sangbum Choi
63
+ - **Funded by:** National Key R&D Program of China (No.2022ZD0118201), Natural Science Foundation of China (No.61972217, 32071459, 62176249, 62006133, 62271465),
64
  and the Shenzhen Medical Research Funds in China (No.
65
  B2302037).
66
+ - **Shared by:** Sangbum Choi
67
+ - **Model type:** [RT-DETR](https://huggingface.co/docs/transformers/main/en/model_doc/rt_detr)
 
68
  - **License:** Apache-2.0
 
69
 
70
+ ### Model Sources
71
 
72
  <!-- Provide the basic links for the model. -->
73
 
74
+ - **HF Docs:** [RT-DETR](https://huggingface.co/docs/transformers/main/en/model_doc/rt_detr)
75
  - **Repository:** https://github.com/lyuwenyu/RT-DETR
76
+ - **Paper:** https://arxiv.org/abs/2304.08069
77
+ - **Demo:** [RT-DETR Tracking](https://huggingface.co/spaces/merve/RT-DETR-tracking-coco)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
78
 
79
  ## How to Get Started with the Model
80
 
81
  Use the code below to get started with the model.
82
 
83
+ ```python
84
  import torch
85
  import requests
86
 
 
127
 
128
  <!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
129
 
130
+ We conduct experiments on COCO and Objects365 datasets, where RT-DETR is trained on COCO train2017 and validated on COCO val2017 dataset.
131
+ We report the standard COCO metrics, including AP (averaged over uniformly sampled IoU thresholds ranging from 0.50-0.95 with a step size of 0.05),
132
+ AP50, AP75, as well as AP at different scales: APS, APM, APL.
 
 
 
133
 
134
+ ### Preprocessing
135
 
136
+ Images are resized to 640x640 pixels and rescaled with `image_mean=[0.485, 0.456, 0.406]` and `image_std=[0.229, 0.224, 0.225]`.
137
 
138
+ ### Training Hyperparameters
139
 
140
  - **Training regime:** <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
141
 
142
  ![image/png](https://cdn-uploads.huggingface.co/production/uploads/6579e0eaa9e58aec614e9d97/E15I9MwZCtwNIms-W8Ra9.png)
143
 
 
 
 
144
 
145
  ## Evaluation
146
 
 
 
 
147
 
148
+ | Model | #Epochs | #Params (M) | GFLOPs | FPS_bs=1 | AP (val) | AP50 (val) | AP75 (val) | AP-s (val) | AP-m (val) | AP-l (val) |
149
+ |----------------------------|---------|-------------|--------|----------|--------|-----------|-----------|----------|----------|----------|
150
+ | RT-DETR-R18 | 72 | 20 | 60.7 | 217 | 46.5 | 63.8 | 50.4 | 28.4 | 49.8 | 63.0 |
151
+ | RT-DETR-R34 | 72 | 31 | 91.0 | 172 | 48.5 | 66.2 | 52.3 | 30.2 | 51.9 | 66.2 |
152
+ | RT-DETR R50 | 72 | 42 | 136 | 108 | 53.1 | 71.3 | 57.7 | 34.8 | 58.0 | 70.0 |
153
+ | RT-DETR R101| 72 | 76 | 259 | 74 | 54.3 | 72.7 | 58.6 | 36.0 | 58.8 | 72.1 |
154
+ | RT-DETR-R18 (Objects 365 pretrained) | 60 | 20 | 61 | 217 | 49.2 | 66.6 | 53.5 | 33.2 | 52.3 | 64.8 |
155
+ | RT-DETR-R50 (Objects 365 pretrained) | 24 | 42 | 136 | 108 | 55.3 | 73.4 | 60.1 | 37.9 | 59.9 | 71.8 |
156
+ | RT-DETR-R101 (Objects 365 pretrained) | 24 | 76 | 259 | 74 | 56.2 | 74.6 | 61.3 | 38.3 | 60.5 | 73.5 |
157
 
 
158
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
159
 
160
  ### Model Architecture and Objective
161
 
162
  ![image/png](https://cdn-uploads.huggingface.co/production/uploads/6579e0eaa9e58aec614e9d97/sdIwTRlHNwPzyBNwHja60.png)
163
 
164
+ Overview of RT-DETR. We feed the features from the last three stages of the backbone into the encoder. The efficient hybrid
165
+ encoder transforms multi-scale features into a sequence of image features through the Attention-based Intra-scale Feature Interaction (AIFI)
166
+ and the CNN-based Cross-scale Feature Fusion (CCFF). Then, the uncertainty-minimal query selection selects a fixed number of encoder
167
+ features to serve as initial object queries for the decoder. Finally, the decoder with auxiliary prediction heads iteratively optimizes object
168
+ queries to generate categories and boxes.
169
 
 
170
 
171
+ ## Citation
 
 
172
 
173
  <!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
174
 
 
185
  }
186
  ```
187
 
188
+ ## Model Card Authors
 
 
 
 
 
 
 
 
189
 
190
+ [Sangbum Choi](https://huggingface.co/danelcsb)
191
+ [Pavel Iakubovskii](https://huggingface.co/qubvel-hf)
192