YOLOv5 tutorial and readme fixes (#339)

* YOLOv5 tutorial and readme fixes

* fix reference to v3

Author: markurtz

integrations/ultralytics-yolov5/README.md

@@ -29,11 +29,12 @@ The techniques include, but are not limited to:
 
 ## Highlights
 
-- Coming soon!
+- Example: [DeepSparse YOLOv5 Inference Example](https://github.com/neuralmagic/deepsparse/tree/main/examples/ultralytics-yolo)
 
 ## Tutorials
 
-- Coming soon!
+- [Sparsifying YOLOv5 Using Recipes](https://github.com/neuralmagic/sparseml/blob/main/integrations/ultralytics-yolov5/tutorials/sparsifying_yolov5_using_recipes.md)
+- [Sparse Transfer Learning With YOLOv5](https://github.com/neuralmagic/sparseml/blob/main/integrations/ultralytics-yolov5/tutorials/yolov5_sparse_transfer_learning.md)
 
 ## Installation
 
@@ -44,6 +45,7 @@ bash setup_integration.sh
 
 The setup_integration.sh file will clone the yolov5 repository with the SparseML integration as a subfolder.
 After the repo has successfully cloned,  all dependencies from the `yolov5/requirements.txt` file will install in your current environment.
+Note, the `yolov5` repository requires Python 3.7 or greater.
 
 ## Quick Tour
 
@@ -84,7 +86,7 @@ The export process is modified such that the quantized and pruned models are cor
 
 For example, the following command can be run from within the yolov5 repository folder to export a trained/sparsified model's checkpoint:
 ```bash
-python models/export.py --weights PATH/TO/weights.pt --img-size 512 512
+python models/export.py --weights PATH/TO/weights.pt --dynamic
 ```
 
 The DeepSparse Engine accepts ONNX formats and is engineered to significantly speed up inference on CPUs for the sparsified models from this integration.

integrations/ultralytics-yolov5/tutorials/images/transfer-learning-wandb-chart.png

@@ -16,7 +16,7 @@ limitations under the License.
 
 # Sparsifying YOLOv5 Using Recipes
 
-This tutorial shows how Neural Magic recipes simplify the sparsification process by encoding the hyperparameters and instructions needed to create highly accurate pruned and pruned-quantized YOLOv5 models, specifically for the S and L versions.
+This tutorial shows how Neural Magic recipes simplify the sparsification process by encoding the hyperparameters and instructions needed to create highly accurate pruned and pruned-quantized YOLOv5 models, specifically for the s and l versions.
 
 ## Overview
 
@@ -55,7 +55,7 @@ Otherwise, setup scripts for both [VOC](https://cs.stanford.edu/~roozbeh/pascal-
 
 1. For this tutorial, run the COCO setup script with the following command from the root of the `yolov5` repository:
     ```bash
-    bash data/scripts/get_voc.sh
+    bash data/scripts/get_coco.sh
     ```
 2. Download and validation of the COCO dataset will begin and takes around 10 minutes to finish.
     The script downloads the COCO dataset into a `coco` folder under the parent directory.
@@ -150,12 +150,12 @@ The table below compares these tradeoffs and shows how to run them on the COCO d
 
     | Recipe Name                                                                                                                                              | Description                                                                                                                     | Train Command                                                                                                                                                                               | COCO mAP@0.5 | Size on Disk | DeepSparse Performance** |
     |----------------------------------------------------------------------------------------------------------------------------------------------------------|---------------------------------------------------------------------------------------------------------------------------------|---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|--------------|--------------|--------------------------|
-    | YOLOv5s Baseline                                                                                                                                         | The baseline, small YOLOv5 model used as the starting point for sparsification.                                                 | ``` python train.py --cfg ../models/yolov5s.yaml --weights "" --data coco.yaml --hyp data/hyp.scratch.yaml ```                                                                              | 0.556        | 154 MB       | --- img/sec              |
-    | [YOLOv5s Pruned](https://github.com/neuralmagic/sparseml/blob/main/integrations/ultralytics-yolov5/recipes/yolov5s.pruned.md)                            | Creates a highly sparse, FP32 YOLOv5s model that recovers close to the baseline model.                                          | ``` python train.py --cfg ../models/yolov5s.yaml --weights PATH_TO_COCO_PRETRAINED_WEIGHTS --data coco.yaml --hyp data/hyp.scratch.yaml --recipe ../recipes/yolov5s.pruned.md ```           | 0.534        | 32.8 MB      | --- img/sec              |
-    | [YOLOv5s Pruned Quantized](https://github.com/neuralmagic/sparseml/blob/main/integrations/ultralytics-yolov5/recipes/yolov5s.pruned_quantized.md)        | Creates a highly sparse, INT8 YOLOv5s model that recovers reasonably close to the baseline model.                               | ``` python train.py --cfg ../models/yolov5s.yaml --weights PATH_TO_COCO_PRETRAINED_WEIGHTS --data coco.yaml --hyp data/hyp.scratch.yaml --recipe ../recipes/yolov5s.pruned_quantized.md ``` | 0.525        | 12.7 MB      | --- img/sec              |
-    | YOLOv5l Baseline                                                                                                                                         | The baseline, large YOLOv5 model used as the starting point for sparsification.                                                 | ``` python train.py --cfg ../models/yolov5l.yaml --weights "" --data coco.yaml --hyp data/hyp.scratch.yaml ```                                                                              | 0.654        | 24.8 MB      | --- img/sec              |
-    | [YOLOv5l Pruned](https://github.com/neuralmagic/sparseml/blob/main/integrations/ultralytics-yolov5/recipes/yolov5l.pruned.md)                            | Creates a highly sparse, FP32 YOLOv5l model that recovers close to the baseline model.                                          | ``` python train.py --cfg ../models/yolov5l.yaml --weights PATH_TO_COCO_PRETRAINED_WEIGHTS --data coco.yaml --hyp data/hyp.scratch.yaml --recipe ../recipes/yolov5l.pruned.md ```           | 0.643        | 8.4 MB       | --- img/sec              |
-    | [YOLOv5l Pruned Quantized](https://github.com/neuralmagic/sparseml/blob/main/integrations/ultralytics-yolov5/recipes/yolov5l.pruned_quantized.md)        | Creates a highly sparse, INT8 YOLOv5l model that recovers reasonably close to the baseline model.                               | ``` python train.py --cfg ../models/yolov5l.yaml --weights PATH_TO_COCO_PRETRAINED_WEIGHTS --data coco.yaml --hyp data/hyp.scratch.yaml --recipe ../recipes/yolov5l.pruned_quantized.md ``` | 0.623        | 3.3 MB       | --- img/sec              |
+    | YOLOv5s Baseline                                                                                                                                         | The baseline, small YOLOv5 model used as the starting point for sparsification.                                                 | ``` python train.py --cfg ../models/yolov5s.yaml --weights "" --data coco.yaml --hyp data/hyp.scratch.yaml ```                                                                              | 0.556        | 154 MB       | 78.2 img/sec             |
+    | [YOLOv5s Pruned](https://github.com/neuralmagic/sparseml/blob/main/integrations/ultralytics-yolov5/recipes/yolov5s.pruned.md)                            | Creates a highly sparse, FP32 YOLOv5s model that recovers close to the baseline model.                                          | ``` python train.py --cfg ../models/yolov5s.yaml --weights PATH_TO_COCO_PRETRAINED_WEIGHTS --data coco.yaml --hyp data/hyp.scratch.yaml --recipe ../recipes/yolov5s.pruned.md ```           | 0.534        | 32.8 MB      | 100.5 img/sec            |
+    | [YOLOv5s Pruned Quantized](https://github.com/neuralmagic/sparseml/blob/main/integrations/ultralytics-yolov5/recipes/yolov5s.pruned_quantized.md)        | Creates a highly sparse, INT8 YOLOv5s model that recovers reasonably close to the baseline model.                               | ``` python train.py --cfg ../models/yolov5s.yaml --weights PATH_TO_COCO_PRETRAINED_WEIGHTS --data coco.yaml --hyp data/hyp.scratch.yaml --recipe ../recipes/yolov5s.pruned_quantized.md ``` | 0.525        | 12.7 MB      | 198.2 img/sec            |
+    | YOLOv5l Baseline                                                                                                                                         | The baseline, large YOLOv5 model used as the starting point for sparsification.                                                 | ``` python train.py --cfg ../models/yolov5l.yaml --weights "" --data coco.yaml --hyp data/hyp.scratch.yaml ```                                                                              | 0.654        | 24.8 MB      | 22.7 img/sec             |
+    | [YOLOv5l Pruned](https://github.com/neuralmagic/sparseml/blob/main/integrations/ultralytics-yolov5/recipes/yolov5l.pruned.md)                            | Creates a highly sparse, FP32 YOLOv5l model that recovers close to the baseline model.                                          | ``` python train.py --cfg ../models/yolov5l.yaml --weights PATH_TO_COCO_PRETRAINED_WEIGHTS --data coco.yaml --hyp data/hyp.scratch.yaml --recipe ../recipes/yolov5l.pruned.md ```           | 0.643        | 8.4 MB       | 40.1 img/sec             |
+    | [YOLOv5l Pruned Quantized](https://github.com/neuralmagic/sparseml/blob/main/integrations/ultralytics-yolov5/recipes/yolov5l.pruned_quantized.md)        | Creates a highly sparse, INT8 YOLOv5l model that recovers reasonably close to the baseline model.                               | ``` python train.py --cfg ../models/yolov5l.yaml --weights PATH_TO_COCO_PRETRAINED_WEIGHTS --data coco.yaml --hyp data/hyp.scratch.yaml --recipe ../recipes/yolov5l.pruned_quantized.md ``` | 0.623        | 3.3 MB       | 98.6 img/sec             |
 
     ** DeepSparse Performance measured on an AWS C5 instance with 24 cores, batch size 64, and 640x640 input with version 1.6 of the DeepSparse Engine.
 
@@ -223,7 +223,7 @@ The [`export.py` script](https://github.com/neuralmagic/yolov5/blob/master/model
 1. Enter the following command to load the PyTorch graph, convert to ONNX, and correct any misformatted pieces of the graph for the pruned and quantized models.
 
     ```bash
-    python models/export.py --weights PATH_TO_SPARSIFIED_WEIGHTS
+    python models/export.py --weights PATH_TO_SPARSIFIED_WEIGHTS --dynamic
     ```
     
     The result is a new file added next to the sparsified checkpoint with a `.onnx` extension:

integrations/ultralytics-yolov5/tutorials/sparsifying_yolov5_using_recipes.md

@@ -47,12 +47,13 @@ For Neural Magic Support, sign up or log in to get help with your questions in o
 
    | Sparsification Type       | Description                                                                                  | COCO mAP@0.5 | Size on Disk | DeepSparse Performance** |
    |---------------------------|----------------------------------------------------------------------------------------------|--------------|--------------|--------------------------|
-   | YOLOv5s Baseline          | The baseline, small YOLOv5 model used as the starting point for sparsification.              | 0.556        | 154 MB       | --- img/sec              |
-   | YOLOv5s Pruned            | A highly sparse, FP32 YOLOv5s model that recovers close to the baseline model.               | 0.534        | 32.8 MB      | --- img/sec              |
-   | YOLOv5s Pruned Quantized  | A highly sparse, INT8 YOLOv5s model that recovers reasonably close to the baseline model.    | 0.525        | 12.7 MB      | --- img/sec              |
-   | YOLOv5l Baseline          | The baseline, large YOLOv5 model used as the starting point for sparsification.              | 0.654        | 24.8 MB      | --- img/sec              |
-   | YOLOv5l Pruned            | A highly sparse, FP32 YOLOv5l model that recovers close to the baseline model.               | 0.643        | 8.4 MB       | --- img/sec              |
-   | YOLOv5l Pruned Quantized  | A highly sparse, INT8 YOLOv5l model that recovers reasonably close to the baseline model.    | 0.623        | 3.3 MB       | --- img/sec              |
+   | YOLOv5s Baseline          | The baseline, small YOLOv5 model used as the starting point for sparsification.              | 0.556        | 154 MB       | 78.2 img/sec             |
+   | YOLOv5s Pruned            | A highly sparse, FP32 YOLOv5s model that recovers close to the baseline model.               | 0.534        | 32.8 MB      | 100.5 img/sec            |
+   | YOLOv5s Pruned Quantized  | A highly sparse, INT8 YOLOv5s model that recovers reasonably close to the baseline model.    | 0.525        | 12.7 MB      | 198.2 img/sec            |
+   | YOLOv5l Baseline          | The baseline, large YOLOv5 model used as the starting point for sparsification.              | 0.654        | 24.8 MB      | 22.7 img/sec             |
+   | YOLOv5l Pruned            | A highly sparse, FP32 YOLOv5l model that recovers close to the baseline model.               | 0.643        | 8.4 MB       | 40.1 img/sec             |
+   | YOLOv5l Pruned Quantized  | A highly sparse, INT8 YOLOv5l model that recovers reasonably close to the baseline model.    | 0.623        | 3.3 MB       | 98.6 img/sec             |
+   
    ** DeepSparse Performance measured on an AWS C5 instance with 24 cores, batch size 64, and 640x640 input with version 1.6 of the DeepSparse Engine.
 
 2) After deciding on which model meets your performance requirements for both speed and accuracy, select the SparseZoo stub associated with that model.
@@ -200,7 +201,7 @@ The [export.py script](https://github.com/neuralmagic/yolov5/blob/master/models/
 
 1) Enter the following command to load the PyTorch graph, convert to ONNX, and correct any misformatted pieces of the graph for the pruned and quantized models.
     ```bash
-    python models/export.py --weights PATH_TO_SPARSIFIED_WEIGHTS --img-size 512 512
+    python models/export.py --weights PATH_TO_SPARSIFIED_WEIGHTS  --dynamic
     ```
     
     The result is a new file added next to the sparsified checkpoint with a `.onnx` extension: