DeepStream 7.1 + Fixes + New model output format

utils/export_goldyolo.py

@@ -0,0 +1,121 @@
+import os
+import onnx
+import torch
+import torch.nn as nn
+
+import yolov6.utils.general as _m
+from yolov6.layers.common import SiLU
+from gold_yolo.switch_tool import switch_to_deploy
+from yolov6.utils.checkpoint import load_checkpoint
+
+
+def _dist2bbox(distance, anchor_points, box_format='xyxy'):
+    lt, rb = torch.split(distance, 2, -1)
+    x1y1 = anchor_points - lt
+    x2y2 = anchor_points + rb
+    bbox = torch.cat([x1y1, x2y2], -1)
+    return bbox
+
+_m.dist2bbox.__code__ = _dist2bbox.__code__
+
+
+class DeepStreamOutput(nn.Module):
+    def __init__(self):
+        super().__init__()
+
+    def forward(self, x):
+        boxes = x[:, :, :4]
+        objectness = x[:, :, 4:5]
+        scores, labels = torch.max(x[:, :, 5:], dim=-1, keepdim=True)
+        scores *= objectness
+        return torch.cat([boxes, scores, labels.to(boxes.dtype)], dim=-1)
+
+
+def gold_yolo_export(weights, device, inplace=True, fuse=True):
+    model = load_checkpoint(weights, map_location=device, inplace=inplace, fuse=fuse)
+    model = switch_to_deploy(model)
+    for layer in model.modules():
+        t = type(layer)
+        if t.__name__ == 'RepVGGBlock':
+            layer.switch_to_deploy()
+    model.eval()
+    for k, m in model.named_modules():
+        if m.__class__.__name__ == 'Conv':
+            if isinstance(m.act, nn.SiLU):
+                m.act = SiLU()
+        elif m.__class__.__name__ == 'Detect':
+            m.inplace = False
+    return model
+
+
+def suppress_warnings():
+    import warnings
+    warnings.filterwarnings('ignore', category=torch.jit.TracerWarning)
+    warnings.filterwarnings('ignore', category=UserWarning)
+    warnings.filterwarnings('ignore', category=DeprecationWarning)
+    warnings.filterwarnings('ignore', category=FutureWarning)
+    warnings.filterwarnings('ignore', category=ResourceWarning)
+
+
+def main(args):
+    suppress_warnings()
+
+    print(f'\nStarting: {args.weights}')
+
+    print('Opening Gold-YOLO model')
+
+    device = torch.device('cpu')
+    model = gold_yolo_export(args.weights, device)
+
+    model = nn.Sequential(model, DeepStreamOutput())
+
+    img_size = args.size * 2 if len(args.size) == 1 else args.size
+
+    onnx_input_im = torch.zeros(args.batch, 3, *img_size).to(device)
+    onnx_output_file = f'{args.weights}.onnx'
+
+    dynamic_axes = {
+        'input': {
+            0: 'batch'
+        },
+        'output': {
+            0: 'batch'
+        }
+    }
+
+    print('Exporting the model to ONNX')
+    torch.onnx.export(
+        model, onnx_input_im, onnx_output_file, verbose=False, opset_version=args.opset, do_constant_folding=True,
+        input_names=['input'], output_names=['output'], dynamic_axes=dynamic_axes if args.dynamic else None
+    )
+
+    if args.simplify:
+        print('Simplifying the ONNX model')
+        import onnxslim
+        model_onnx = onnx.load(onnx_output_file)
+        model_onnx = onnxslim.slim(model_onnx)
+        onnx.save(model_onnx, onnx_output_file)
+
+    print(f'Done: {onnx_output_file}\n')
+
+
+def parse_args():
+    import argparse
+    parser = argparse.ArgumentParser(description='DeepStream Gold-YOLO conversion')
+    parser.add_argument('-w', '--weights', required=True, help='Input weights (.pt) file path (required)')
+    parser.add_argument('-s', '--size', nargs='+', type=int, default=[640], help='Inference size [H,W] (default [640])')
+    parser.add_argument('--opset', type=int, default=13, help='ONNX opset version')
+    parser.add_argument('--simplify', action='store_true', help='ONNX simplify model')
+    parser.add_argument('--dynamic', action='store_true', help='Dynamic batch-size')
+    parser.add_argument('--batch', type=int, default=1, help='Static batch-size')
+    args = parser.parse_args()
+    if not os.path.isfile(args.weights):
+        raise SystemExit('Invalid weights file')
+    if args.dynamic and args.batch > 1:
+        raise SystemExit('Cannot set dynamic batch-size and static batch-size at same time')
+    return args
+
+
+if __name__ == '__main__':
+    args = parse_args()
+    main(args)