TDA4VM: Edgeai-benchmark yolov5 model compilation error.

Sefa BEKER

Part Number: TDA4VM

Hi,

I have the Jacinto J7 EVM kit.

https://github.com/TexasInstruments/edgeai-yolov5 Using the relevant repo, I'm using my special data set to train the yolov5s6. After I do the onnx conversion, I have two files. "best.onnx" and "best.prototxt"

The contents of the "best.prototxt" file looks like this.

name: "yolo_v3"
tidl_yolo {
  yolo_param {
    input: "370"
    anchor_width: 19.0
    anchor_width: 44.0
    anchor_width: 38.0
    anchor_height: 27.0
    anchor_height: 40.0
    anchor_height: 94.0
  }
  yolo_param {
    input: "426"
    anchor_width: 96.0
    anchor_width: 86.0
    anchor_width: 180.0
    anchor_height: 68.0
    anchor_height: 152.0
    anchor_height: 137.0
  }
  yolo_param {
    input: "482"
    anchor_width: 140.0
    anchor_width: 303.0
    anchor_width: 238.0
    anchor_height: 301.0
    anchor_height: 264.0
    anchor_height: 542.0
  }
  yolo_param {
    input: "538"
    anchor_width: 436.0
    anchor_width: 739.0
    anchor_width: 925.0
    anchor_height: 615.0
    anchor_height: 380.0
    anchor_height: 792.0
  }
  detection_output_param {
    num_classes: 36
    share_location: true
    background_label_id: -1
    nms_param {
      nms_threshold: 0.65
      top_k: 30000
    }
    code_type: CODE_TYPE_YOLO_V5
    keep_top_k: 300
    confidence_threshold: 0.005
  }
  name: "yolo_v3"
  in_width: 640
  in_height: 640
  output: "detections"
}

https://github.com/TexasInstruments/edgeai-benchmark I want to compile my custom model to run on TDA4VM using the corresponding repo.

For this, I set my pipeline_config settings in benchmark_custom.py as follows.

'imagedet-best': dict(
            task_type='detection',
            calibration_dataset=imagedet_calib_dataset,
            input_dataset=imagedet_val_dataset,
            preprocess=preproc_transforms.get_transform_onnx((640,640), (640,640), resize_with_pad=[True], backend='cv2'),
            session=sessions.ONNXRTSession(**onnx_session_cfg,
                runtime_options=utils.dict_update(settings.runtime_options_onnx_np2(),
                                    {'object_detection:meta_arch_type': 6,
                                     'object_detection:meta_layers_names_list':'/home/sefau18/edgeai-modelzoo/models/vision/detection/coco/bests6v2/best.prototxt',
                                     'advanced_options:output_feature_16bit_names_list':'370, 426, 482, 538'
                                     }),
                model_path='/home/sefau18/edgeai-modelzoo/models/vision/detection/coco/bests6v2/best.onnx',
            postprocess=postproc_transforms.get_transform_detection_yolov5_onnx(squeeze_axis=None, normalized_detections=False, resize_with_pad=True, formatter=postprocess.DetectionBoxSL2BoxLS()),
            metric=dict(label_offset_pred=datasets.coco_det_label_offset_90to90()),
            model_info=dict(metric_reference={'accuracy_ap[.5:.95]%':45.0})
        ))

I am sharing the error I got.

(benchmark) sefau18@ubuntu:~/edgeai-benchmark$ ./run_custom_pc.sh 
Entering: ./work_dirs/modelartifacts/8bits/cl-3420_tvmdlr_imagenet1k_gluoncv-mxnet_resnet50_v1d-symbol_json.tar.gz.link
run_set_target_device.sh: line 59: cd: ./work_dirs/modelartifacts/8bits/cl-3420_tvmdlr_imagenet1k_gluoncv-mxnet_resnet50_v1d-symbol_json.tar.gz.link/artifacts: Not a directory
Entering: ./work_dirs/modelartifacts/8bits/cl-3410_tvmdlr_imagenet1k_gluoncv-mxnet_mobilenetv2_1.0-symbol_json.tar.gz.link
run_set_target_device.sh: line 59: cd: ./work_dirs/modelartifacts/8bits/cl-3410_tvmdlr_imagenet1k_gluoncv-mxnet_mobilenetv2_1.0-symbol_json.tar.gz.link/artifacts: Not a directory
Entering: ./work_dirs/modelartifacts/8bits/cl-3420_tvmdlr_imagenet1k_gluoncv-mxnet_resnet50_v1d-symbol_json.tar.gz.link.link.link
run_set_target_device.sh: line 59: cd: ./work_dirs/modelartifacts/8bits/cl-3420_tvmdlr_imagenet1k_gluoncv-mxnet_resnet50_v1d-symbol_json.tar.gz.link.link.link/artifacts: Not a directory
Entering: ./work_dirs/modelartifacts/8bits/od-5040_tvmdlr_coco_gluoncv-mxnet_ssd_512_mobilenet1.0_coco-symbol_json.tar.gz.link.link
run_set_target_device.sh: line 59: cd: ./work_dirs/modelartifacts/8bits/od-5040_tvmdlr_coco_gluoncv-mxnet_ssd_512_mobilenet1.0_coco-symbol_json.tar.gz.link.link/artifacts: Not a directory
Entering: ./work_dirs/modelartifacts/8bits/cl-3410_tvmdlr_imagenet1k_gluoncv-mxnet_mobilenetv2_1.0-symbol_json.tar.gz.link.link.link
run_set_target_device.sh: line 59: cd: ./work_dirs/modelartifacts/8bits/cl-3410_tvmdlr_imagenet1k_gluoncv-mxnet_mobilenetv2_1.0-symbol_json.tar.gz.link.link.link/artifacts: Not a directory
Entering: ./work_dirs/modelartifacts/8bits/od-5040_tvmdlr_coco_gluoncv-mxnet_ssd_512_mobilenet1.0_coco-symbol_json.tar.gz.link.link.link
run_set_target_device.sh: line 59: cd: ./work_dirs/modelartifacts/8bits/od-5040_tvmdlr_coco_gluoncv-mxnet_ssd_512_mobilenet1.0_coco-symbol_json.tar.gz.link.link.link/artifacts: Not a directory
Entering: ./work_dirs/modelartifacts/8bits/ss-5818_tvmdlr_ti-robokit_edgeai-tv_deeplabv3plus_mobilenetv2_tv_edgeailite_robokit-zed1hd_768x432_qat-p2_onnx.tar.gz.link.link.link
run_set_target_device.sh: line 59: cd: ./work_dirs/modelartifacts/8bits/ss-5818_tvmdlr_ti-robokit_edgeai-tv_deeplabv3plus_mobilenetv2_tv_edgeailite_robokit-zed1hd_768x432_qat-p2_onnx.tar.gz.link.link.link/artifacts: Not a directory
Entering: ./work_dirs/modelartifacts/8bits/ss-5720_tvmdlr_cocoseg21_edgeai-tv_fpn_aspp_regnetx800mf_edgeailite_512x512_20210405_onnx.tar.gz.link.link
run_set_target_device.sh: line 59: cd: ./work_dirs/modelartifacts/8bits/ss-5720_tvmdlr_cocoseg21_edgeai-tv_fpn_aspp_regnetx800mf_edgeailite_512x512_20210405_onnx.tar.gz.link.link/artifacts: Not a directory
Entering: ./work_dirs/modelartifacts/8bits/cl-3480_tvmdlr_imagenet1k_gluoncv-mxnet_hrnet_w18_small_v2_c-symbol_json.tar.gz.link.link
run_set_target_device.sh: line 59: cd: ./work_dirs/modelartifacts/8bits/cl-3480_tvmdlr_imagenet1k_gluoncv-mxnet_hrnet_w18_small_v2_c-symbol_json.tar.gz.link.link/artifacts: Not a directory
Entering: ./work_dirs/modelartifacts/8bits/od-5030_tvmdlr_coco_gluoncv-mxnet_ssd_512_resnet50_v1_coco-symbol_json.tar.gz.link.link
run_set_target_device.sh: line 59: cd: ./work_dirs/modelartifacts/8bits/od-5030_tvmdlr_coco_gluoncv-mxnet_ssd_512_resnet50_v1_coco-symbol_json.tar.gz.link.link/artifacts: Not a directory
Entering: ./work_dirs/modelartifacts/8bits/cl-3430_tvmdlr_imagenet1k_gluoncv-mxnet_xception-symbol_json.tar.gz.link
run_set_target_device.sh: line 59: cd: ./work_dirs/modelartifacts/8bits/cl-3430_tvmdlr_imagenet1k_gluoncv-mxnet_xception-symbol_json.tar.gz.link/artifacts: Not a directory
Entering: ./work_dirs/modelartifacts/8bits/od-5030_tvmdlr_coco_gluoncv-mxnet_ssd_512_resnet50_v1_coco-symbol_json.tar.gz.link
run_set_target_device.sh: line 59: cd: ./work_dirs/modelartifacts/8bits/od-5030_tvmdlr_coco_gluoncv-mxnet_ssd_512_resnet50_v1_coco-symbol_json.tar.gz.link/artifacts: Not a directory
Entering: ./work_dirs/modelartifacts/8bits/od-5030_tvmdlr_coco_gluoncv-mxnet_ssd_512_resnet50_v1_coco-symbol_json.tar.gz.link.link.link
run_set_target_device.sh: line 59: cd: ./work_dirs/modelartifacts/8bits/od-5030_tvmdlr_coco_gluoncv-mxnet_ssd_512_resnet50_v1_coco-symbol_json.tar.gz.link.link.link/artifacts: Not a directory
Entering: ./work_dirs/modelartifacts/8bits/cl-3430_tvmdlr_imagenet1k_gluoncv-mxnet_xception-symbol_json.tar.gz.link.link.link
run_set_target_device.sh: line 59: cd: ./work_dirs/modelartifacts/8bits/cl-3430_tvmdlr_imagenet1k_gluoncv-mxnet_xception-symbol_json.tar.gz.link.link.link/artifacts: Not a directory
Entering: ./work_dirs/modelartifacts/8bits/ss-5720_tvmdlr_cocoseg21_edgeai-tv_fpn_aspp_regnetx800mf_edgeailite_512x512_20210405_onnx.tar.gz.link
run_set_target_device.sh: line 59: cd: ./work_dirs/modelartifacts/8bits/ss-5720_tvmdlr_cocoseg21_edgeai-tv_fpn_aspp_regnetx800mf_edgeailite_512x512_20210405_onnx.tar.gz.link/artifacts: Not a directory
Entering: ./work_dirs/modelartifacts/8bits/od-5040_tvmdlr_coco_gluoncv-mxnet_ssd_512_mobilenet1.0_coco-symbol_json.tar.gz.link
run_set_target_device.sh: line 59: cd: ./work_dirs/modelartifacts/8bits/od-5040_tvmdlr_coco_gluoncv-mxnet_ssd_512_mobilenet1.0_coco-symbol_json.tar.gz.link/artifacts: Not a directory
Entering: ./work_dirs/modelartifacts/8bits/ss-5818_tvmdlr_ti-robokit_edgeai-tv_deeplabv3plus_mobilenetv2_tv_edgeailite_robokit-zed1hd_768x432_qat-p2_onnx.tar.gz.link
run_set_target_device.sh: line 59: cd: ./work_dirs/modelartifacts/8bits/ss-5818_tvmdlr_ti-robokit_edgeai-tv_deeplabv3plus_mobilenetv2_tv_edgeailite_robokit-zed1hd_768x432_qat-p2_onnx.tar.gz.link/artifacts: Not a directory
Entering: ./work_dirs/modelartifacts/8bits/od-5020_tvmdlr_coco_gluoncv-mxnet_yolo3_mobilenet1.0_coco-symbol_json.tar.gz.link.link
run_set_target_device.sh: line 59: cd: ./work_dirs/modelartifacts/8bits/od-5020_tvmdlr_coco_gluoncv-mxnet_yolo3_mobilenet1.0_coco-symbol_json.tar.gz.link.link/artifacts: Not a directory
Entering: ./work_dirs/modelartifacts/8bits/cl-3420_tvmdlr_imagenet1k_gluoncv-mxnet_resnet50_v1d-symbol_json.tar.gz.link.link
run_set_target_device.sh: line 59: cd: ./work_dirs/modelartifacts/8bits/cl-3420_tvmdlr_imagenet1k_gluoncv-mxnet_resnet50_v1d-symbol_json.tar.gz.link.link/artifacts: Not a directory
Entering: ./work_dirs/modelartifacts/8bits/od-5020_tvmdlr_coco_gluoncv-mxnet_yolo3_mobilenet1.0_coco-symbol_json.tar.gz.link.link.link
run_set_target_device.sh: line 59: cd: ./work_dirs/modelartifacts/8bits/od-5020_tvmdlr_coco_gluoncv-mxnet_yolo3_mobilenet1.0_coco-symbol_json.tar.gz.link.link.link/artifacts: Not a directory
Entering: ./work_dirs/modelartifacts/8bits/cl-3480_tvmdlr_imagenet1k_gluoncv-mxnet_hrnet_w18_small_v2_c-symbol_json.tar.gz.link
run_set_target_device.sh: line 59: cd: ./work_dirs/modelartifacts/8bits/cl-3480_tvmdlr_imagenet1k_gluoncv-mxnet_hrnet_w18_small_v2_c-symbol_json.tar.gz.link/artifacts: Not a directory
Entering: ./work_dirs/modelartifacts/8bits/cl-3430_tvmdlr_imagenet1k_gluoncv-mxnet_xception-symbol_json.tar.gz.link.link
run_set_target_device.sh: line 59: cd: ./work_dirs/modelartifacts/8bits/cl-3430_tvmdlr_imagenet1k_gluoncv-mxnet_xception-symbol_json.tar.gz.link.link/artifacts: Not a directory
Entering: ./work_dirs/modelartifacts/8bits/od-5020_tvmdlr_coco_gluoncv-mxnet_yolo3_mobilenet1.0_coco-symbol_json.tar.gz.link
run_set_target_device.sh: line 59: cd: ./work_dirs/modelartifacts/8bits/od-5020_tvmdlr_coco_gluoncv-mxnet_yolo3_mobilenet1.0_coco-symbol_json.tar.gz.link/artifacts: Not a directory
Entering: ./work_dirs/modelartifacts/8bits/ss-5720_tvmdlr_cocoseg21_edgeai-tv_fpn_aspp_regnetx800mf_edgeailite_512x512_20210405_onnx.tar.gz.link.link.link
run_set_target_device.sh: line 59: cd: ./work_dirs/modelartifacts/8bits/ss-5720_tvmdlr_cocoseg21_edgeai-tv_fpn_aspp_regnetx800mf_edgeailite_512x512_20210405_onnx.tar.gz.link.link.link/artifacts: Not a directory
Entering: ./work_dirs/modelartifacts/8bits/ss-5818_tvmdlr_ti-robokit_edgeai-tv_deeplabv3plus_mobilenetv2_tv_edgeailite_robokit-zed1hd_768x432_qat-p2_onnx.tar.gz.link.link
run_set_target_device.sh: line 59: cd: ./work_dirs/modelartifacts/8bits/ss-5818_tvmdlr_ti-robokit_edgeai-tv_deeplabv3plus_mobilenetv2_tv_edgeailite_robokit-zed1hd_768x432_qat-p2_onnx.tar.gz.link.link/artifacts: Not a directory
Entering: ./work_dirs/modelartifacts/8bits/cl-3410_tvmdlr_imagenet1k_gluoncv-mxnet_mobilenetv2_1.0-symbol_json.tar.gz.link.link
run_set_target_device.sh: line 59: cd: ./work_dirs/modelartifacts/8bits/cl-3410_tvmdlr_imagenet1k_gluoncv-mxnet_mobilenetv2_1.0-symbol_json.tar.gz.link.link/artifacts: Not a directory
Entering: ./work_dirs/modelartifacts/8bits/cl-3480_tvmdlr_imagenet1k_gluoncv-mxnet_hrnet_w18_small_v2_c-symbol_json.tar.gz.link.link.link
run_set_target_device.sh: line 59: cd: ./work_dirs/modelartifacts/8bits/cl-3480_tvmdlr_imagenet1k_gluoncv-mxnet_hrnet_w18_small_v2_c-symbol_json.tar.gz.link.link.link/artifacts: Not a directory
TIDL_TOOLS_PATH=/home/sefau18/edgeai-benchmark/tidl_tools
LD_LIBRARY_PATH=/home/sefau18/edgeai-benchmark/tidl_tools
PYTHONPATH=:
===================================================================
work_dir = ./work_dirs/modelartifacts/8bits
loading annotations into memory...
Done (t=0.30s)
creating index...
index created!
loading annotations into memory...
Done (t=0.37s)
creating index...
index created!
configs to run: ['imagedet-20_onnxrt_coco_bests6v2_best_onnx']
number of configs: 1
TASKS                                                       |          |     0% 0/1| [< ]
INFO:20220727-224225: starting process on parallel_device - 0   0%|          || 0/1 [00:00<?, ?it/s
INFO:20220727-224229: model_path - /home/sefau18/edgeai-modelzoo/models/vision/detection/coco/bests6v2/best.onnx
INFO:20220727-224229: model_file - /home/sefau18/edgeai-benchmark/work_dirs/modelartifacts/8bits/imagedet-20_onnxrt_coco_bests6v2_best_onnx/model/best.onnx

INFO:20220727-224229: running - imagedet-20_onnxrt_coco_bests6v2_best_onnx
INFO:20220727-224229: pipeline_config - {'task_type': 'detection', 'calibration_dataset': <jai_benchmark.datasets.coco_det.COCODetection object at 0x7ff6d0597c88>, 'input_dataset': <jai_benchmark.datasets.coco_det.COCODetection object at 0x7ff6d89e7780>, 'preprocess': <jai_benchmark.preprocess.PreProcessTransforms object at 0x7ff6d89e74e0>, 'session': <jai_benchmark.sessions.onnxrt_session.ONNXRTSession object at 0x7ff6ca7ed588>}
INFO:20220727-224229: import  - imagedet-20_onnxrt_coco_bests6v2_best_onnxTIDL Meta PipeLine (Proto) File  : /home/sefau18/edgeai-benchmark/work_dirs/modelartifacts/8bits/imagedet-20_onnxrt_coco_bests6v2_best_onnx/model/best.prototxt  
yolo_v3
yolo_v3
Number of OD backbone nodes = 195 
Size of odBackboneNodeIds = 195 

Preliminary subgraphs created = 1 
Final number of subgraphs created are : 1, - Offloaded Nodes - 298, Total Nodes - 298 
TIDL Meta PipeLine (Proto) File  : /home/sefau18/edgeai-benchmark/work_dirs/modelartifacts/8bits/imagedet-20_onnxrt_coco_bests6v2_best_onnx/model/best.prototxt  
yolo_v3
yolo_v3
****************************************************
**   All the Input Tensor Dimensions has to be greater then Zero 
**   DIM Error - For Tensor 196, Dim 1 is -1
****************************************************

Editing the prototxt file according to the original doesn't make any sense.

Is there a step-by-step documentation on how to compile a custom yolov5 model and perform inference using "edgeai apps" on the TDA4VM?

Thanks already for your help.

kumar.desappan

Manu Mathew

over 3 years ago

0 Manu Mathew over 3 years ago

TI__Genius 11506 points

Can you zip your onnx model and prototxt and attach here. We can try it out at our end.

0 Sefa BEKER over 3 years ago in reply to Manu Mathew

Prodigy 35 points

Hi Mathew,

Thank you for the answer.
https://github.com/TexasInstruments/edgeai-yolov5

Below are the files you requested for two different trainings I made using the same data set.

1- $ python3 train.py --data data.yaml --cfg yolov5s6.yaml --weights 'yolov5s6.pt' --batch-size 128

yolov5s6.yaml file

only nc number edited

# Parameters
nc: 36  # number of classes
depth_multiple: 0.33  # model depth multiple
width_multiple: 0.50  # layer channel multiple
anchors:
  - [ 19,27,  44,40,  38,94 ]  # P3/8
  - [ 96,68,  86,152,  180,137 ]  # P4/16
  - [ 140,301,  303,264,  238,542 ]  # P5/32
  - [ 436,615,  739,380,  925,792 ]  # P6/64

# YOLOv5 backbone
backbone:
  # [from, number, module, args]
  [ [ -1, 1, Focus, [ 64, 3 ] ],  # 0-P1/2
    [ -1, 1, Conv, [ 128, 3, 2 ] ],  # 1-P2/4
    [ -1, 3, C3, [ 128 ] ],
    [ -1, 1, Conv, [ 256, 3, 2 ] ],  # 3-P3/8
    [ -1, 9, C3, [ 256 ] ],
    [ -1, 1, Conv, [ 512, 3, 2 ] ],  # 5-P4/16
    [ -1, 9, C3, [ 512 ] ],
    [ -1, 1, Conv, [ 768, 3, 2 ] ],  # 7-P5/32
    [ -1, 3, C3, [ 768 ] ],
    [ -1, 1, Conv, [ 1024, 3, 2 ] ],  # 9-P6/64
    [ -1, 1, SPP, [ 1024, [ 3, 5, 7 ] ] ],
    [ -1, 3, C3, [ 1024, False ] ],  # 11
  ]

# YOLOv5 head
head:
  [ [ -1, 1, Conv, [ 768, 1, 1 ] ],
    [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ],
    [ [ -1, 8 ], 1, Concat, [ 1 ] ],  # cat backbone P5
    [ -1, 3, C3, [ 768, False ] ],  # 15

    [ -1, 1, Conv, [ 512, 1, 1 ] ],
    [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ],
    [ [ -1, 6 ], 1, Concat, [ 1 ] ],  # cat backbone P4
    [ -1, 3, C3, [ 512, False ] ],  # 19

    [ -1, 1, Conv, [ 256, 1, 1 ] ],
    [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ],
    [ [ -1, 4 ], 1, Concat, [ 1 ] ],  # cat backbone P3
    [ -1, 3, C3, [ 256, False ] ],  # 23 (P3/8-small)

    [ -1, 1, Conv, [ 256, 3, 2 ] ],
    [ [ -1, 20 ], 1, Concat, [ 1 ] ],  # cat head P4
    [ -1, 3, C3, [ 512, False ] ],  # 26 (P4/16-medium)

    [ -1, 1, Conv, [ 512, 3, 2 ] ],
    [ [ -1, 16 ], 1, Concat, [ 1 ] ],  # cat head P5
    [ -1, 3, C3, [ 768, False ] ],  # 29 (P5/32-large)

    [ -1, 1, Conv, [ 768, 3, 2 ] ],
    [ [ -1, 12 ], 1, Concat, [ 1 ] ],  # cat head P6
    [ -1, 3, C3, [ 1024, False ] ],  # 32 (P6/64-xlarge)

    [ [ 23, 26, 29, 32 ], 1, Detect, [ nc, anchors ] ],  # Detect(P3, P4, P5, P6)
  ]

best.zip

2- $ python3 train.py --data data.yaml --cfg yolov5s6.yaml --weights '' --batch-size 128

Training started with no initial weight given and change to anchors in yaml file. Anchor values were created automatically by the software during conversion.

yolov5s6.yaml file

only anchors edited

# Parameters
nc: 36  # number of classes
depth_multiple: 0.33  # model depth multiple
width_multiple: 0.50  # layer channel multiple
anchors: 3
# YOLOv5 backbone
backbone:
  # [from, number, module, args]
  [ [ -1, 1, Focus, [ 64, 3 ] ],  # 0-P1/2
    [ -1, 1, Conv, [ 128, 3, 2 ] ],  # 1-P2/4
    [ -1, 3, C3, [ 128 ] ],
    [ -1, 1, Conv, [ 256, 3, 2 ] ],  # 3-P3/8
    [ -1, 9, C3, [ 256 ] ],
    [ -1, 1, Conv, [ 512, 3, 2 ] ],  # 5-P4/16
    [ -1, 9, C3, [ 512 ] ],
    [ -1, 1, Conv, [ 768, 3, 2 ] ],  # 7-P5/32
    [ -1, 3, C3, [ 768 ] ],
    [ -1, 1, Conv, [ 1024, 3, 2 ] ],  # 9-P6/64
    [ -1, 1, SPP, [ 1024, [ 3, 5, 7 ] ] ],
    [ -1, 3, C3, [ 1024, False ] ],  # 11
  ]

# YOLOv5 head
head:
  [ [ -1, 1, Conv, [ 768, 1, 1 ] ],
    [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ],
    [ [ -1, 8 ], 1, Concat, [ 1 ] ],  # cat backbone P5
    [ -1, 3, C3, [ 768, False ] ],  # 15

    [ -1, 1, Conv, [ 512, 1, 1 ] ],
    [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ],
    [ [ -1, 6 ], 1, Concat, [ 1 ] ],  # cat backbone P4
    [ -1, 3, C3, [ 512, False ] ],  # 19

    [ -1, 1, Conv, [ 256, 1, 1 ] ],
    [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ],
    [ [ -1, 4 ], 1, Concat, [ 1 ] ],  # cat backbone P3
    [ -1, 3, C3, [ 256, False ] ],  # 23 (P3/8-small)

    [ -1, 1, Conv, [ 256, 3, 2 ] ],
    [ [ -1, 20 ], 1, Concat, [ 1 ] ],  # cat head P4
    [ -1, 3, C3, [ 512, False ] ],  # 26 (P4/16-medium)

    [ -1, 1, Conv, [ 512, 3, 2 ] ],
    [ [ -1, 16 ], 1, Concat, [ 1 ] ],  # cat head P5
    [ -1, 3, C3, [ 768, False ] ],  # 29 (P5/32-large)

    [ -1, 1, Conv, [ 768, 3, 2 ] ],
    [ [ -1, 12 ], 1, Concat, [ 1 ] ],  # cat head P6
    [ -1, 3, C3, [ 1024, False ] ],  # 32 (P6/64-xlarge)

    [ [ 23, 26, 29, 32 ], 1, Detect, [ nc, anchors ] ],  # Detect(P3, P4, P5, P6)
  ]

2best.zip

Thanks for the help.