i use card 3090. It seems only support cuda 11.2.

Hi, thank you very much for your graceful work.
I would like to ask if you have any plans to release the Faster R-CNN pre-trained model? I would appreciate it if you could do this, it would help me a lot.
Thank you very much!

Hi, thank you for your nice work.
Can you tell me which detectron2 version have been used?
I have a problem when evaluate the model(training by myself or downloading from the link)

[05/06 11:12:29] d2.evaluation.coco_evaluation WARNING: No predictions from the model!
[05/06 11:12:29] d2.engine.defaults INFO: Evaluation results for coco_2017_val in csv format:
[05/06 11:12:29] d2.evaluation.testing INFO: copypaste: Task: bbox
[05/06 11:12:29] d2.evaluation.testing INFO: copypaste: AP,AP50,AP75,APs,APm,APl
[05/06 11:12:29] d2.evaluation.testing INFO: copypaste: nan,nan,nan,nan,nan,nan

I want to make sure whether it is a version issue, thanks~

Thank you for your nice project!
I called your swint-Detectron2 code to train my own data set. I used the network configuration of retinanet. No error was reported during the training process,iteration=200000 AP50=0.7 or so, and the test set is only AP50=0.5 or so. I have changed pixel_mean according to my dataset.
I would like to know what other parameters should be paid attention to in training, which may be helpful for my training!

Thanks for your nice project.

What do the note mean?

the weight i download from the “model” which in the table, that weight can't be used directly?

the downloaded weight also have to convert to d2 format?

I also want to know that the metric ”boxmap” is evaluated on the which datasets.

Thanks for your works.

I want to know what is timm in swin_transformer.py, line 11

你好，我想问下model中swintransformer好像并不是级联的结构

输出的四个stage好像是互不影响的，请问这样的原因是什么呢?

Thank you for your nice project! i trained my own data with your backbone then i want to inference my result.but it shows that it Cannot find field 'pred_masks' in the given Instances! .i use mask_rcnn_swint_T_FPN_3x.yaml as my config.

faster_rcnn训练ap为nan ，点击错了，希望重新打开

Hi,
My code work with Detectron 0.3 and I want change it's backbone from Resnet to SwinT,
you said your code work with Detectron 0.6, Do you have any idea about solve this problem?

Thanks

Hi, thanks for your excellent work. I wonder if Swin-Transformer supports FRCN C-4? How to write the config? Thanks!

请问，RGB 和BGR以及mean，std是怎么对应的

Hi,
Thanks for sharing your Code. I tried to run convert_to_d2.py file to create model file. But i have an error and I guess it's because of torch version. Can you help me about step of running and please tell which version torch i should to use.

Thank you

The metrics of the results are nan.

The train_net.py as follow:

#!/usr/bin/env python

Copyright (c) Facebook, Inc. and its affiliates.

"""
Detection Training Script.
This scripts reads a given config file and runs the training or evaluation.
It is an entry point that is made to train standard models in detectron2.
In order to let one script support training of many models,
this script contains logic that are specific to these built-in models and therefore
may not be suitable for your own project.
For example, your research project perhaps only needs a single "evaluator".
Therefore, we recommend you to use detectron2 as an library and take
this file as an example of how to use the library.
You may want to write your own script with your datasets and other customizations.
"""
import itertools
import logging
import os
from collections import OrderedDict
import torch

import detectron2.utils.comm as comm
from detectron2.checkpoint import DetectionCheckpointer
from detectron2.config import get_cfg
from detectron2.data import MetadataCatalog
from detectron2.engine import DefaultTrainer, default_argument_parser, default_setup, hooks, launch
from detectron2.evaluation import (
CityscapesInstanceEvaluator,
CityscapesSemSegEvaluator,
COCOEvaluator,
COCOPanopticEvaluator,
DatasetEvaluators,
LVISEvaluator,
PascalVOCDetectionEvaluator,
SemSegEvaluator,
verify_results,
)
from detectron2.modeling import GeneralizedRCNNWithTTA
from detectron2.solver.build import maybe_add_gradient_clipping, get_default_optimizer_params

from swint import add_swint_config
import pig_dataset

class Trainer(DefaultTrainer):
"""
We use the "DefaultTrainer" which contains pre-defined default logic for
standard training workflow. They may not work for you, especially if you
are working on a new research project. In that case you can write your
own training loop. You can use "tools/plain_train_net.py" as an example.
"""

@classmethod
def build_evaluator(cls, cfg, dataset_name, output_folder=None):
    """
    Create evaluator(s) for a given dataset.
    This uses the special metadata "evaluator_type" associated with each builtin dataset.
    For your own dataset, you can simply create an evaluator manually in your
    script and do not have to worry about the hacky if-else logic here.
    """
    if output_folder is None:
        output_folder = os.path.join(cfg.OUTPUT_DIR, "inference")
    evaluator_list = []
    evaluator_type = MetadataCatalog.get(dataset_name).evaluator_type
    if evaluator_type in ["sem_seg", "coco_panoptic_seg"]:
        evaluator_list.append(
            SemSegEvaluator(
                dataset_name,
                distributed=True,
                num_classes=cfg.MODEL.SEM_SEG_HEAD.NUM_CLASSES,
                ignore_label=cfg.MODEL.SEM_SEG_HEAD.IGNORE_VALUE,
                output_dir=output_folder,
            )
        )
    if evaluator_type in ["coco", "coco_panoptic_seg"]:
        evaluator_list.append(COCOEvaluator(dataset_name, cfg, True, output_folder))
    if evaluator_type == "coco_panoptic_seg":
        evaluator_list.append(COCOPanopticEvaluator(dataset_name, output_folder))
    if evaluator_type == "cityscapes_instance":
        assert (
            torch.cuda.device_count() >= comm.get_rank()
        ), "CityscapesEvaluator currently do not work with multiple machines."
        return CityscapesInstanceEvaluator(dataset_name)
    if evaluator_type == "cityscapes_sem_seg":
        assert (
            torch.cuda.device_count() >= comm.get_rank()
        ), "CityscapesEvaluator currently do not work with multiple machines."
        return CityscapesSemSegEvaluator(dataset_name)
    elif evaluator_type == "pascal_voc":
        return PascalVOCDetectionEvaluator(dataset_name)
    elif evaluator_type == "lvis":
        return LVISEvaluator(dataset_name, cfg, True, output_folder)
    if len(evaluator_list) == 0:
        raise NotImplementedError(
            "no Evaluator for the dataset {} with the type {}".format(
                dataset_name, evaluator_type
            )
        )
    elif len(evaluator_list) == 1:
        return evaluator_list[0]
    return DatasetEvaluators(evaluator_list)

@classmethod
def test_with_TTA(cls, cfg, model):
    logger = logging.getLogger("detectron2.trainer")
    # In the end of training, run an evaluation with TTA
    # Only support some R-CNN models.
    logger.info("Running inference with test-time augmentation ...")
    model = GeneralizedRCNNWithTTA(cfg, model)
    evaluators = [
        cls.build_evaluator(
            cfg, name, output_folder=os.path.join(cfg.OUTPUT_DIR, "inference_TTA")
        )
        for name in cfg.DATASETS.TEST
    ]
    res = cls.test(cfg, model, evaluators)
    res = OrderedDict({k + "_TTA": v for k, v in res.items()})
    return res

@classmethod
def build_optimizer(cls, cfg, model):
    params = get_default_optimizer_params(
        model,
        base_lr=cfg.SOLVER.BASE_LR,
        weight_decay=cfg.SOLVER.WEIGHT_DECAY,
        weight_decay_norm=cfg.SOLVER.WEIGHT_DECAY_NORM,
        bias_lr_factor=cfg.SOLVER.BIAS_LR_FACTOR,
        weight_decay_bias=cfg.SOLVER.WEIGHT_DECAY_BIAS,
    )

    def maybe_add_full_model_gradient_clipping(optim):  # optim: the optimizer class
        # detectron2 doesn't have full model gradient clipping now
        clip_norm_val = cfg.SOLVER.CLIP_GRADIENTS.CLIP_VALUE
        enable = (
            cfg.SOLVER.CLIP_GRADIENTS.ENABLED
            and cfg.SOLVER.CLIP_GRADIENTS.CLIP_TYPE == "full_model"
            and clip_norm_val > 0.0
        )

        class FullModelGradientClippingOptimizer(optim):
            def step(self, closure=None):
                all_params = itertools.chain(*[x["params"] for x in self.param_groups])
                torch.nn.utils.clip_grad_norm_(all_params, clip_norm_val)
                super().step(closure=closure)

        return FullModelGradientClippingOptimizer if enable else optim

    optimizer_type = cfg.SOLVER.OPTIMIZER
    if optimizer_type == "SGD":
        optimizer = maybe_add_gradient_clipping(torch.optim.SGD)(
            params, cfg.SOLVER.BASE_LR, momentum=cfg.SOLVER.MOMENTUM,
            nesterov=cfg.SOLVER.NESTEROV,
            weight_decay=cfg.SOLVER.WEIGHT_DECAY,
        )
    elif optimizer_type == "AdamW":
        optimizer = maybe_add_full_model_gradient_clipping(torch.optim.AdamW)(
            params, cfg.SOLVER.BASE_LR, betas=(0.9, 0.999),
            weight_decay=cfg.SOLVER.WEIGHT_DECAY,
        )

    else:
        raise NotImplementedError("no optimizer type {optimizer_type}")
    return optimizer

def setup(args):
"""
Create configs and perform basic setups.
"""
cfg = get_cfg()
add_swint_config(cfg)
args.config_file = "./configs/SwinT/retinanet_swint_T_FPN_3x.yaml"####
cfg.merge_from_file(args.config_file)
cfg.MODEL.WEIGHTS = "weights/retinanet_swint_S_3x.pth"###
cfg.DATASETS.TRAIN = ("pig_coco_train",)
cfg.DATASETS.TEST = ("pig_coco_test", )
cfg.merge_from_list(args.opts)
cfg.freeze()
default_setup(cfg, args)
return cfg

def main(args):
cfg = setup(args)
args.eval_only = True

if args.eval_only:
    model = Trainer.build_model(cfg)
    DetectionCheckpointer(model, save_dir=cfg.OUTPUT_DIR).resume_or_load(
        cfg.MODEL.WEIGHTS, resume=args.resume
    )
    res = Trainer.test(cfg, model)
    if cfg.TEST.AUG.ENABLED:
        res.update(Trainer.test_with_TTA(cfg, model))
    if comm.is_main_process():
        verify_results(cfg, res)
    return res

"""
If you'd like to do anything fancier than the standard training logic,
consider writing your own training loop (see plain_train_net.py) or
subclassing the trainer.
"""
trainer = Trainer(cfg)
trainer.resume_or_load(resume=args.resume)
if cfg.TEST.AUG.ENABLED:
    trainer.register_hooks(
        [hooks.EvalHook(0, lambda: trainer.test_with_TTA(cfg, trainer.model))]
    )
return trainer.train()

if name == "main":
args = default_argument_parser().parse_args()
print("Command Line Args:", args)
launch(
main,
args.num_gpus,
num_machines=args.num_machines,
machine_rank=args.machine_rank,
dist_url=args.dist_url,
args=(args,),
)

=================================================================================================
The output as follow:

/home/server/anaconda3/envs/swin_d2/bin/python /home/server/SwinT_detectron2/train_net.py
Command Line Args: Namespace(config_file='', dist_url='tcp://127.0.0.1:50152', eval_only=False, machine_rank=0, num_gpus=1, num_machines=1, opts=[], resume=False)
Loading config ./configs/SwinT/../Base-RetinaNet.yaml with yaml.unsafe_load. Your machine may be at risk if the file contains malicious content.
[08/26 12:26:16 detectron2]: Rank of current process: 0. World size: 1
[08/26 12:26:17 detectron2]: Environment info:

sys.platform linux
Python 3.7.11 (default, Jul 27 2021, 14:32:16) [GCC 7.5.0]
numpy 1.19.2
detectron2 0.5 @/home/server/anaconda3/envs/swin_d2/lib/python3.7/site-packages/detectron2
Compiler GCC 7.3
CUDA compiler CUDA 11.1
detectron2 arch flags 3.7, 5.0, 5.2, 6.0, 6.1, 7.0, 7.5, 8.0, 8.6
DETECTRON2_ENV_MODULE
PyTorch 1.9.0 @/home/server/anaconda3/envs/swin_d2/lib/python3.7/site-packages/torch
PyTorch debug build False
GPU available Yes
GPU 0 NVIDIA GeForce RTX 3090 (arch=8.6)
CUDA_HOME /usr/local/cuda
Pillow 8.3.1
torchvision 0.10.0 @/home/server/anaconda3/envs/swin_d2/lib/python3.7/site-packages/torchvision
torchvision arch flags 3.5, 5.0, 6.0, 7.0, 7.5, 8.0, 8.6
fvcore 0.1.5.post20210804
iopath 0.1.8
cv2 4.4.0

PyTorch built with:

• GCC 7.3
• C++ Version: 201402
• Intel(R) Math Kernel Library Version 2020.0.2 Product Build 20200624 for Intel(R) 64 architecture applications
• Intel(R) MKL-DNN v2.1.2 (Git Hash 98be7e8afa711dc9b66c8ff3504129cb82013cdb)
• OpenMP 201511 (a.k.a. OpenMP 4.5)
• NNPACK is enabled
• CPU capability usage: AVX2
• CUDA Runtime 11.1
• NVCC architecture flags: -gencode;arch=compute_37,code=sm_37;-gencode;arch=compute_50,code=sm_50;-gencode;arch=compute_60,code=sm_60;-gencode;arch=compute_61,code=sm_61;-gencode;arch=compute_70,code=sm_70;-gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80;-gencode;arch=compute_86,code=sm_86;-gencode;arch=compute_37,code=compute_37
• CuDNN 8.0.5
• Magma 2.5.2
• Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, CUDA_VERSION=11.1, CUDNN_VERSION=8.0.5, CXX_COMPILER=/opt/rh/devtoolset-7/root/usr/bin/c++, CXX_FLAGS= -Wno-deprecated -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -fopenmp -DNDEBUG -DUSE_KINETO -DUSE_FBGEMM -DUSE_QNNPACK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -O2 -fPIC -Wno-narrowing -Wall -Wextra -Werror=return-type -Wno-missing-field-initializers -Wno-type-limits -Wno-array-bounds -Wno-unknown-pragmas -Wno-sign-compare -Wno-unused-parameter -Wno-unused-variable -Wno-unused-function -Wno-unused-result -Wno-unused-local-typedefs -Wno-strict-overflow -Wno-strict-aliasing -Wno-error=deprecated-declarations -Wno-stringop-overflow -Wno-psabi -Wno-error=pedantic -Wno-error=redundant-decls -Wno-error=old-style-cast -fdiagnostics-color=always -faligned-new -Wno-unused-but-set-variable -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Wno-stringop-overflow, LAPACK_INFO=mkl, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, PERF_WITH_AVX512=1, TORCH_VERSION=1.9.0, USE_CUDA=ON, USE_CUDNN=ON, USE_EXCEPTION_PTR=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=ON, USE_NNPACK=ON, USE_OPENMP=ON,

[08/26 12:26:17 detectron2]: Command line arguments: Namespace(config_file='./configs/SwinT/retinanet_swint_T_FPN_3x.yaml', dist_url='tcp://127.0.0.1:50152', eval_only=False, machine_rank=0, num_gpus=1, num_machines=1, opts=[], resume=False)
[08/26 12:26:17 detectron2]: Contents of args.config_file=./configs/SwinT/retinanet_swint_T_FPN_3x.yaml:
BASE: "../Base-RetinaNet.yaml"
MODEL:
WEIGHTS: "weights/retinanet_swint_S_3x.pth"
PIXEL_MEAN: [123.675, 116.28, 103.53] # use RGB [103.530, 116.280, 123.675]
PIXEL_STD: [58.395, 57.12, 57.375] #[57.375, 57.120, 58.395] # I use the dafault config [1.0, 1.0, 1.0] and BGR format, that is a mistake
RESNETS:
DEPTH: 50
BACKBONE:
NAME: "build_retinanet_swint_fpn_backbone"
SWINT:
OUT_FEATURES: ["stage3", "stage4", "stage5"]
FPN:
IN_FEATURES: ["stage3", "stage4", "stage5"]
INPUT:
FORMAT: "RGB"
SOLVER:
STEPS: (210000, 250000)
MAX_ITER: 270000
WEIGHT_DECAY: 0.05
BASE_LR: 0.0001
AMP:
ENABLED: True
TEST:
EVAL_PERIOD: 30000

DATASETS:
TRAIN: ("coco_2017_train",)
TEST: ("coco_2017_val",)

[08/26 12:26:17 detectron2]: Running with full config:
CUDNN_BENCHMARK: false
DATALOADER:
ASPECT_RATIO_GROUPING: true
FILTER_EMPTY_ANNOTATIONS: true
NUM_WORKERS: 4
REPEAT_THRESHOLD: 0.0
SAMPLER_TRAIN: TrainingSampler
DATASETS:
PRECOMPUTED_PROPOSAL_TOPK_TEST: 1000
PRECOMPUTED_PROPOSAL_TOPK_TRAIN: 2000
PROPOSAL_FILES_TEST: []
PROPOSAL_FILES_TRAIN: []
TEST:

• pig_coco_test
  TRAIN:
• pig_coco_train
  GLOBAL:
  HACK: 1.0
  INPUT:
  CROP:
  ENABLED: false
  SIZE:
  • 0.9
  • 0.9
    TYPE: relative_range
    FORMAT: RGB
    MASK_FORMAT: polygon
    MAX_SIZE_TEST: 1333
    MAX_SIZE_TRAIN: 1333
    MIN_SIZE_TEST: 800
    MIN_SIZE_TRAIN:
• 640
• 672
• 704
• 736
• 768
• 800
  MIN_SIZE_TRAIN_SAMPLING: choice
  RANDOM_FLIP: horizontal
  MODEL:
  ANCHOR_GENERATOR:
  ANGLES:
  • • -90
    • 0
    • 90
      ASPECT_RATIOS:
  • • 0.5
    • 1.0
    • 2.0
      NAME: DefaultAnchorGenerator
      OFFSET: 0.0
      SIZES:
  • • 32
    • 40.31747359663594
    • 50.79683366298238
  • • 64
    • 80.63494719327188
    • 101.59366732596476
  • • 128
    • 161.26989438654377
    • 203.18733465192952
  • • 256
    • 322.53978877308754
    • 406.37466930385904
  • • 512
    • 645.0795775461751
    • 812.7493386077181
      BACKBONE:
      FREEZE_AT: -1
      NAME: build_retinanet_swint_fpn_backbone
      DEVICE: cuda
      FPN:
      FUSE_TYPE: sum
      IN_FEATURES:
  • stage3
  • stage4
  • stage5
    NORM: ''
    OUT_CHANNELS: 256
    TOP_LEVELS: 2
    KEYPOINT_ON: false
    LOAD_PROPOSALS: false
    MASK_ON: false
    META_ARCHITECTURE: RetinaNet
    PANOPTIC_FPN:
    COMBINE:
    ENABLED: true
    INSTANCES_CONFIDENCE_THRESH: 0.5
    OVERLAP_THRESH: 0.5
    STUFF_AREA_LIMIT: 4096
    INSTANCE_LOSS_WEIGHT: 1.0
    PIXEL_MEAN:
• 123.675
• 116.28
• 103.53
  PIXEL_STD:
• 58.395
• 57.12
• 57.375
  PROPOSAL_GENERATOR:
  MIN_SIZE: 0
  NAME: RPN
  RESNETS:
  DEFORM_MODULATED: false
  DEFORM_NUM_GROUPS: 1
  DEFORM_ON_PER_STAGE:
  • false
  • false
  • false
  • false
    DEPTH: 50
    NORM: FrozenBN
    NUM_GROUPS: 1
    OUT_FEATURES:
  • res3
  • res4
  • res5
    RES2_OUT_CHANNELS: 256
    RES5_DILATION: 1
    STEM_OUT_CHANNELS: 64
    STRIDE_IN_1X1: true
    WIDTH_PER_GROUP: 64
    RETINANET:
    BBOX_REG_LOSS_TYPE: smooth_l1
    BBOX_REG_WEIGHTS: &id001
  • 1.0
  • 1.0
  • 1.0
  • 1.0
    FOCAL_LOSS_ALPHA: 0.25
    FOCAL_LOSS_GAMMA: 2.0
    IN_FEATURES:
  • p3
  • p4
  • p5
  • p6
  • p7
    IOU_LABELS:
  • 0
  • -1
  • 1
    IOU_THRESHOLDS:
  • 0.4
  • 0.5
    NMS_THRESH_TEST: 0.5
    NORM: ''
    NUM_CLASSES: 80
    NUM_CONVS: 4
    PRIOR_PROB: 0.01
    SCORE_THRESH_TEST: 0.05
    SMOOTH_L1_LOSS_BETA: 0.0
    TOPK_CANDIDATES_TEST: 1000
    ROI_BOX_CASCADE_HEAD:
    BBOX_REG_WEIGHTS:
  • • 10.0
    • 10.0
    • 5.0
    • 5.0
  • • 20.0
    • 20.0
    • 10.0
    • 10.0
  • • 30.0
    • 30.0
    • 15.0
    • 15.0
      IOUS:
  • 0.5
  • 0.6
  • 0.7
    ROI_BOX_HEAD:
    BBOX_REG_LOSS_TYPE: smooth_l1
    BBOX_REG_LOSS_WEIGHT: 1.0
    BBOX_REG_WEIGHTS:
  • 10.0
  • 10.0
  • 5.0
  • 5.0
    CLS_AGNOSTIC_BBOX_REG: false
    CONV_DIM: 256
    FC_DIM: 1024
    NAME: ''
    NORM: ''
    NUM_CONV: 0
    NUM_FC: 0
    POOLER_RESOLUTION: 14
    POOLER_SAMPLING_RATIO: 0
    POOLER_TYPE: ROIAlignV2
    SMOOTH_L1_BETA: 0.0
    TRAIN_ON_PRED_BOXES: false
    ROI_HEADS:
    BATCH_SIZE_PER_IMAGE: 512
    IN_FEATURES:
  • res4
    IOU_LABELS:
  • 0
  • 1
    IOU_THRESHOLDS:
  • 0.5
    NAME: Res5ROIHeads
    NMS_THRESH_TEST: 0.5
    NUM_CLASSES: 80
    POSITIVE_FRACTION: 0.25
    PROPOSAL_APPEND_GT: true
    SCORE_THRESH_TEST: 0.05
    ROI_KEYPOINT_HEAD:
    CONV_DIMS:
  • 512
  • 512
  • 512
  • 512
  • 512
  • 512
  • 512
  • 512
    LOSS_WEIGHT: 1.0
    MIN_KEYPOINTS_PER_IMAGE: 1
    NAME: KRCNNConvDeconvUpsampleHead
    NORMALIZE_LOSS_BY_VISIBLE_KEYPOINTS: true
    NUM_KEYPOINTS: 17
    POOLER_RESOLUTION: 14
    POOLER_SAMPLING_RATIO: 0
    POOLER_TYPE: ROIAlignV2
    ROI_MASK_HEAD:
    CLS_AGNOSTIC_MASK: false
    CONV_DIM: 256
    NAME: MaskRCNNConvUpsampleHead
    NORM: ''
    NUM_CONV: 0
    POOLER_RESOLUTION: 14
    POOLER_SAMPLING_RATIO: 0
    POOLER_TYPE: ROIAlignV2
    RPN:
    BATCH_SIZE_PER_IMAGE: 256
    BBOX_REG_LOSS_TYPE: smooth_l1
    BBOX_REG_LOSS_WEIGHT: 1.0
    BBOX_REG_WEIGHTS: *id001
    BOUNDARY_THRESH: -1
    CONV_DIMS:
  • -1
    HEAD_NAME: StandardRPNHead
    IN_FEATURES:
  • res4
    IOU_LABELS:
  • 0
  • -1
  • 1
    IOU_THRESHOLDS:
  • 0.3
  • 0.7
    LOSS_WEIGHT: 1.0
    NMS_THRESH: 0.7
    POSITIVE_FRACTION: 0.5
    POST_NMS_TOPK_TEST: 1000
    POST_NMS_TOPK_TRAIN: 2000
    PRE_NMS_TOPK_TEST: 6000
    PRE_NMS_TOPK_TRAIN: 12000
    SMOOTH_L1_BETA: 0.0
    SEM_SEG_HEAD:
    COMMON_STRIDE: 4
    CONVS_DIM: 128
    IGNORE_VALUE: 255
    IN_FEATURES:
  • p2
  • p3
  • p4
  • p5
    LOSS_WEIGHT: 1.0
    NAME: SemSegFPNHead
    NORM: GN
    NUM_CLASSES: 54
    SWINT:
    APE: false
    DEPTHS:
  • 2
  • 2
  • 6
  • 2
    DROP_PATH_RATE: 0.2
    EMBED_DIM: 96
    MLP_RATIO: 4
    NUM_HEADS:
  • 3
  • 6
  • 12
  • 24
    OUT_FEATURES:
  • stage3
  • stage4
  • stage5
    WINDOW_SIZE: 7
    WEIGHTS: weights/retinanet_swint_S_3x.pth
    OUTPUT_DIR: ./output
    SEED: -1
    SOLVER:
    AMP:
    ENABLED: true
    BASE_LR: 0.0001
    BIAS_LR_FACTOR: 1.0
    CHECKPOINT_PERIOD: 5000
    CLIP_GRADIENTS:
    CLIP_TYPE: value
    CLIP_VALUE: 1.0
    ENABLED: false
    NORM_TYPE: 2.0
    GAMMA: 0.1
    IMS_PER_BATCH: 16
    LR_SCHEDULER_NAME: WarmupMultiStepLR
    MAX_ITER: 270000
    MOMENTUM: 0.9
    NESTEROV: false
    OPTIMIZER: AdamW
    REFERENCE_WORLD_SIZE: 0
    STEPS:
• 210000
• 250000
  WARMUP_FACTOR: 0.001
  WARMUP_ITERS: 1000
  WARMUP_METHOD: linear
  WEIGHT_DECAY: 0.05
  WEIGHT_DECAY_BIAS: 0.0001
  WEIGHT_DECAY_NORM: 0.0
  TEST:
  AUG:
  ENABLED: false
  FLIP: true
  MAX_SIZE: 4000
  MIN_SIZES:
  • 400
  • 500
  • 600
  • 700
  • 800
  • 900
  • 1000
  • 1100
  • 1200
    DETECTIONS_PER_IMAGE: 100
    EVAL_PERIOD: 30000
    EXPECTED_RESULTS: []
    KEYPOINT_OKS_SIGMAS: []
    PRECISE_BN:
    ENABLED: false
    NUM_ITER: 200
    VERSION: 2
    VIS_PERIOD: 0

[08/26 12:26:17 detectron2]: Full config saved to ./output/config.yaml
[08/26 12:26:17 d2.utils.env]: Using a generated random seed 17194009
[08/26 12:26:18 d2.engine.defaults]: Model:
RetinaNet(
(backbone): FPN(
(fpn_lateral3): Conv2d(192, 256, kernel_size=(1, 1), stride=(1, 1))
(fpn_output3): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(fpn_lateral4): Conv2d(384, 256, kernel_size=(1, 1), stride=(1, 1))
(fpn_output4): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(fpn_lateral5): Conv2d(768, 256, kernel_size=(1, 1), stride=(1, 1))
(fpn_output5): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(top_block): LastLevelP6P7(
(p6): Conv2d(256, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
(p7): Conv2d(256, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
)
(bottom_up): SwinTransformer(
(patch_embed): PatchEmbed(
(proj): Conv2d(3, 96, kernel_size=(4, 4), stride=(4, 4))
(norm): LayerNorm((96,), eps=1e-05, elementwise_affine=True)
)
(pos_drop): Dropout(p=0.0, inplace=False)
(layers): ModuleList(
(0): BasicLayer(
(blocks): ModuleList(
(0): SwinTransformerBlock(
(norm1): LayerNorm((96,), eps=1e-05, elementwise_affine=True)
(attn): WindowAttention(
(qkv): Linear(in_features=96, out_features=288, bias=True)
(attn_drop): Dropout(p=0.0, inplace=False)
(proj): Linear(in_features=96, out_features=96, bias=True)
(proj_drop): Dropout(p=0.0, inplace=False)
(softmax): Softmax(dim=-1)
)
(drop_path): Identity()
(norm2): LayerNorm((96,), eps=1e-05, elementwise_affine=True)
(mlp): Mlp(
(fc1): Linear(in_features=96, out_features=384, bias=True)
(act): GELU()
(fc2): Linear(in_features=384, out_features=96, bias=True)
(drop): Dropout(p=0.0, inplace=False)
)
)
(1): SwinTransformerBlock(
(norm1): LayerNorm((96,), eps=1e-05, elementwise_affine=True)
(attn): WindowAttention(
(qkv): Linear(in_features=96, out_features=288, bias=True)
(attn_drop): Dropout(p=0.0, inplace=False)
(proj): Linear(in_features=96, out_features=96, bias=True)
(proj_drop): Dropout(p=0.0, inplace=False)
(softmax): Softmax(dim=-1)
)
(drop_path): DropPath()
(norm2): LayerNorm((96,), eps=1e-05, elementwise_affine=True)
(mlp): Mlp(
(fc1): Linear(in_features=96, out_features=384, bias=True)
(act): GELU()
(fc2): Linear(in_features=384, out_features=96, bias=True)
(drop): Dropout(p=0.0, inplace=False)
)
)
)
(downsample): PatchMerging(
(reduction): Linear(in_features=384, out_features=192, bias=False)
(norm): LayerNorm((384,), eps=1e-05, elementwise_affine=True)
)
)
(1): BasicLayer(
(blocks): ModuleList(
(0): SwinTransformerBlock(
(norm1): LayerNorm((192,), eps=1e-05, elementwise_affine=True)
(attn): WindowAttention(
(qkv): Linear(in_features=192, out_features=576, bias=True)
(attn_drop): Dropout(p=0.0, inplace=False)
(proj): Linear(in_features=192, out_features=192, bias=True)
(proj_drop): Dropout(p=0.0, inplace=False)
(softmax): Softmax(dim=-1)
)
(drop_path): DropPath()
(norm2): LayerNorm((192,), eps=1e-05, elementwise_affine=True)
(mlp): Mlp(
(fc1): Linear(in_features=192, out_features=768, bias=True)
(act): GELU()
(fc2): Linear(in_features=768, out_features=192, bias=True)
(drop): Dropout(p=0.0, inplace=False)
)
)
(1): SwinTransformerBlock(
(norm1): LayerNorm((192,), eps=1e-05, elementwise_affine=True)
(attn): WindowAttention(
(qkv): Linear(in_features=192, out_features=576, bias=True)
(attn_drop): Dropout(p=0.0, inplace=False)
(proj): Linear(in_features=192, out_features=192, bias=True)
(proj_drop): Dropout(p=0.0, inplace=False)
(softmax): Softmax(dim=-1)
)
(drop_path): DropPath()
(norm2): LayerNorm((192,), eps=1e-05, elementwise_affine=True)
(mlp): Mlp(
(fc1): Linear(in_features=192, out_features=768, bias=True)
(act): GELU()
(fc2): Linear(in_features=768, out_features=192, bias=True)
(drop): Dropout(p=0.0, inplace=False)
)
)
)
(downsample): PatchMerging(
(reduction): Linear(in_features=768, out_features=384, bias=False)
(norm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
)
)
(2): BasicLayer(
(blocks): ModuleList(
(0): SwinTransformerBlock(
(norm1): LayerNorm((384,), eps=1e-05, elementwise_affine=True)
(attn): WindowAttention(
(qkv): Linear(in_features=384, out_features=1152, bias=True)
(attn_drop): Dropout(p=0.0, inplace=False)
(proj): Linear(in_features=384, out_features=384, bias=True)
(proj_drop): Dropout(p=0.0, inplace=False)
(softmax): Softmax(dim=-1)
)
(drop_path): DropPath()
(norm2): LayerNorm((384,), eps=1e-05, elementwise_affine=True)
(mlp): Mlp(
(fc1): Linear(in_features=384, out_features=1536, bias=True)
(act): GELU()
(fc2): Linear(in_features=1536, out_features=384, bias=True)
(drop): Dropout(p=0.0, inplace=False)
)
)
(1): SwinTransformerBlock(
(norm1): LayerNorm((384,), eps=1e-05, elementwise_affine=True)
(attn): WindowAttention(
(qkv): Linear(in_features=384, out_features=1152, bias=True)
(attn_drop): Dropout(p=0.0, inplace=False)
(proj): Linear(in_features=384, out_features=384, bias=True)
(proj_drop): Dropout(p=0.0, inplace=False)
(softmax): Softmax(dim=-1)
)
(drop_path): DropPath()
(norm2): LayerNorm((384,), eps=1e-05, elementwise_affine=True)
(mlp): Mlp(
(fc1): Linear(in_features=384, out_features=1536, bias=True)
(act): GELU()
(fc2): Linear(in_features=1536, out_features=384, bias=True)
(drop): Dropout(p=0.0, inplace=False)
)
)
(2): SwinTransformerBlock(
(norm1): LayerNorm((384,), eps=1e-05, elementwise_affine=True)
(attn): WindowAttention(
(qkv): Linear(in_features=384, out_features=1152, bias=True)
(attn_drop): Dropout(p=0.0, inplace=False)
(proj): Linear(in_features=384, out_features=384, bias=True)
(proj_drop): Dropout(p=0.0, inplace=False)
(softmax): Softmax(dim=-1)
)
(drop_path): DropPath()
(norm2): LayerNorm((384,), eps=1e-05, elementwise_affine=True)
(mlp): Mlp(
(fc1): Linear(in_features=384, out_features=1536, bias=True)
(act): GELU()
(fc2): Linear(in_features=1536, out_features=384, bias=True)
(drop): Dropout(p=0.0, inplace=False)
)
)
(3): SwinTransformerBlock(
(norm1): LayerNorm((384,), eps=1e-05, elementwise_affine=True)
(attn): WindowAttention(
(qkv): Linear(in_features=384, out_features=1152, bias=True)
(attn_drop): Dropout(p=0.0, inplace=False)
(proj): Linear(in_features=384, out_features=384, bias=True)
(proj_drop): Dropout(p=0.0, inplace=False)
(softmax): Softmax(dim=-1)
)
(drop_path): DropPath()
(norm2): LayerNorm((384,), eps=1e-05, elementwise_affine=True)
(mlp): Mlp(
(fc1): Linear(in_features=384, out_features=1536, bias=True)
(act): GELU()
(fc2): Linear(in_features=1536, out_features=384, bias=True)
(drop): Dropout(p=0.0, inplace=False)
)
)
(4): SwinTransformerBlock(
(norm1): LayerNorm((384,), eps=1e-05, elementwise_affine=True)
(attn): WindowAttention(
(qkv): Linear(in_features=384, out_features=1152, bias=True)
(attn_drop): Dropout(p=0.0, inplace=False)
(proj): Linear(in_features=384, out_features=384, bias=True)
(proj_drop): Dropout(p=0.0, inplace=False)
(softmax): Softmax(dim=-1)
)
(drop_path): DropPath()
(norm2): LayerNorm((384,), eps=1e-05, elementwise_affine=True)
(mlp): Mlp(
(fc1): Linear(in_features=384, out_features=1536, bias=True)
(act): GELU()
(fc2): Linear(in_features=1536, out_features=384, bias=True)
(drop): Dropout(p=0.0, inplace=False)
)
)
(5): SwinTransformerBlock(
(norm1): LayerNorm((384,), eps=1e-05, elementwise_affine=True)
(attn): WindowAttention(
(qkv): Linear(in_features=384, out_features=1152, bias=True)
(attn_drop): Dropout(p=0.0, inplace=False)
(proj): Linear(in_features=384, out_features=384, bias=True)
(proj_drop): Dropout(p=0.0, inplace=False)
(softmax): Softmax(dim=-1)
)
(drop_path): DropPath()
(norm2): LayerNorm((384,), eps=1e-05, elementwise_affine=True)
(mlp): Mlp(
(fc1): Linear(in_features=384, out_features=1536, bias=True)
(act): GELU()
(fc2): Linear(in_features=1536, out_features=384, bias=True)
(drop): Dropout(p=0.0, inplace=False)
)
)
)
(downsample): PatchMerging(
(reduction): Linear(in_features=1536, out_features=768, bias=False)
(norm): LayerNorm((1536,), eps=1e-05, elementwise_affine=True)
)
)
(3): BasicLayer(
(blocks): ModuleList(
(0): SwinTransformerBlock(
(norm1): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
(attn): WindowAttention(
(qkv): Linear(in_features=768, out_features=2304, bias=True)
(attn_drop): Dropout(p=0.0, inplace=False)
(proj): Linear(in_features=768, out_features=768, bias=True)
(proj_drop): Dropout(p=0.0, inplace=False)
(softmax): Softmax(dim=-1)
)
(drop_path): DropPath()
(norm2): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
(mlp): Mlp(
(fc1): Linear(in_features=768, out_features=3072, bias=True)
(act): GELU()
(fc2): Linear(in_features=3072, out_features=768, bias=True)
(drop): Dropout(p=0.0, inplace=False)
)
)
(1): SwinTransformerBlock(
(norm1): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
(attn): WindowAttention(
(qkv): Linear(in_features=768, out_features=2304, bias=True)
(attn_drop): Dropout(p=0.0, inplace=False)
(proj): Linear(in_features=768, out_features=768, bias=True)
(proj_drop): Dropout(p=0.0, inplace=False)
(softmax): Softmax(dim=-1)
)
(drop_path): DropPath()
(norm2): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
(mlp): Mlp(
(fc1): Linear(in_features=768, out_features=3072, bias=True)
(act): GELU()
(fc2): Linear(in_features=3072, out_features=768, bias=True)
(drop): Dropout(p=0.0, inplace=False)
)
)
)
)
)
(norm1): LayerNorm((192,), eps=1e-05, elementwise_affine=True)
(norm2): LayerNorm((384,), eps=1e-05, elementwise_affine=True)
(norm3): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
)
)
(head): RetinaNetHead(
(cls_subnet): Sequential(
(0): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): ReLU()
(2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(3): ReLU()
(4): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(5): ReLU()
(6): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(7): ReLU()
)
(bbox_subnet): Sequential(
(0): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): ReLU()
(2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(3): ReLU()
(4): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(5): ReLU()
(6): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(7): ReLU()
)
(cls_score): Conv2d(256, 720, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(bbox_pred): Conv2d(256, 36, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
)
(anchor_generator): DefaultAnchorGenerator(
(cell_anchors): BufferList()
)
)
[08/26 12:26:18 fvcore.common.checkpoint]: [Checkpointer] Loading from weights/retinanet_swint_S_3x.pth ...
WARNING [08/26 12:26:19 fvcore.common.checkpoint]: The checkpoint state_dict contains keys that are not used by the model:
pixel_mean
pixel_std
anchor_generator.cell_anchors.{0, 1, 2, 3, 4}
[08/26 12:26:19 d2.data.datasets.coco]: Loaded 5000 images in COCO format from ./datasets/coco2017/annotations/instances_val2017.json
[08/26 12:26:19 d2.data.build]: Distribution of instances among all 80 categories:

Process finished with exit code 0

Hello, I would like to ask why there are parameters related to resnet in mask_rcnn_swint_T_FPN_3x.yaml. I think that the swin transformer replaces the original resnet as the backbone network, and there should be no resnet related configuration in the yaml file. Looking forward to your answer!

Hi, I am wondering how many GPUs you have used for reproducing the results. Is this model resource-intensive?

请问你的代码是可以单独运行的，还是需要将你的代码再添加进detectron2 的官方代码中再运行？

Hi,
Thanks for your work.
I have a question about this project.
Do we need to change RGB, PIXEL_MEAN, PIXEL_STD of the configuration, to keep consistency with the original SwinTransformer?

Hi, thank you for your nice work. I can confirm it works and yield nice results even if it is trained on a custom dataset.

Can you please add a license to the repository? E.g., MIT. It is necessary to use the source code correctly and not 'steal' it. Thank you:)

请问目前这种实现方式训练后的模型，是否可以trace 成可以直接被libtorch 直接调用的pt model 呢

Hi, I am using SwinT with mask R-CNN and everything compile and was easy to setup.
However, when I use the pretrained weights (mask_rcnn_swint_T_coco17.pth) to finetuned the model on my custom dataset, I can't seem to acheive over 35 APbb. With ResNet I usually acheive over 60 APbb. Is the pretrained weight the one you got on coco or imagenet? Any tips on what could cause the AP difference?

Thank you

I want to fine tune DiT for object detection (text, diagrams detection only) etc for my own dataset. Been searching through the web for quite some time but could not find anything on fine tuning a Transformers backbone for object detection.

1. I know how to fine tune Detectron 2 for an object detection task with the default given configuration yaml files using Faster RCNN / Masked RCNN models with Resnet or any other backbone CNN models but I don't know how to do it with Transformers models.

2. This github issues for DETR for custom backbone describes how to change the backbone as the author said that you can use ANY models from timm library and since there are almost 890 models present but unfortunately, not DiT.

3. DiT is also present as a HuggingFace model and supports Feature Extraction as BeitFeatureExtractor.from_pretrained("microsoft/dit-large") so I think it could be used as a backbone but I found nothing on this one either.

I tried changing the code on how to train DETR on custom data by replacing code in Cell 8,

#feature_extractor = DetrFeatureExtractor.from_pretrained("facebook/detr-resnet-50")

feature_extractor = BeitFeatureExtractor.from_pretrained("microsoft/dit-large")

but while running the code for Cell 11,

from torch.utils.data import DataLoader

def collate_fn(batch):
  pixel_values = [item[0] for item in batch]
  encoding = feature_extractor.pad_and_create_pixel_mask(pixel_values, return_tensors="pt")
  labels = [item[1] for item in batch]
  batch = {}
  batch['pixel_values'] = encoding['pixel_values']
  batch['pixel_mask'] = encoding['pixel_mask']
  batch['labels'] = labels
  return batch

train_dataloader = DataLoader(train_dataset, collate_fn=collate_fn, batch_size=4, shuffle=True)
val_dataloader = DataLoader(val_dataset, collate_fn=collate_fn, batch_size=2)
batch = next(iter(train_dataloader))

it gave me error as:

---------------------------------------------------------------------------
KeyError                                  Traceback (most recent call last)
<ipython-input-11-446d81c845dd> in <module>
     13 train_dataloader = DataLoader(train_dataset, collate_fn=collate_fn, batch_size=4, shuffle=True)
     14 val_dataloader = DataLoader(val_dataset, collate_fn=collate_fn, batch_size=2)
---> 15 batch = next(iter(train_dataloader))

5 frames
/usr/local/lib/python3.7/dist-packages/transformers/feature_extraction_utils.py in __getitem__(self, item)
     85         """
     86         if isinstance(item, str):
---> 87             return self.data[item]
     88         else:
     89             raise KeyError("Indexing with integers is not available when using Python based feature extractors")

KeyError: 'labels'

Can someone please help me with the problem t hand?

ANY architecture like Faster RCNN, DETR etc and ANY repo or platform like Detectron 2, PaddleDetection, MMDetection, HuggingFace, EfficientDet would do.

Thank you for the awesome project. Can the Mask R-CNN model be trained for instance segmentation?

cfg = get_cfg()
add_swint_config(cfg)
cfg.MODEL.WEIGHTS = "/content/drive/MyDrive/ybigta1/mask_rcnn_swint_T_coco17.pth"

I have wrote as such, but the training does not give mask mAP.

Thanks for your works!

I want to know whether the Images size can be random resized?

Hey,

Thanks a lot for the repo. It is fairly easy to use and get it to work.

However, I am having trouble figuring out how can I set the number of training cycles or the number of epochs for training. I am only using 275 training images but the interface shows more than a day of training time which is - I assume - abnormally high.

Thanks

from detectron2.utils.visualizer import ColorMode
import matplotlib.pyplot as plt
import matplotlib.patches as patches
import json
import pandas as pd
from random import randint
import torch, torchvision

import detectron2
from detectron2.utils.logger import setup_logger
setup_logger()

import numpy as np
import cv2
import random
import matplotlib.pyplot as plt

from detectron2.engine import DefaultPredictor
from detectron2.config import get_cfg
from detectron2.utils.visualizer import Visualizer
from detectron2.data import MetadataCatalog

from swint import add_swint_config

from detectron2.data.datasets import register_coco_instances
register_coco_instances("data_train", {}, "/home/sangjoon/detectron2/sangjoon/for_newthing_0331/white_train.json", "/home/sangjoon/detectron2/sangjoon/white_train2020")
register_coco_instances("data_val", {}, "/home/sangjoon/detectron2/sangjoon/for_newthing_0331/white_test.json", "/home/sangjoon/detectron2/sangjoon/white_test2020")

import os
import numpy as np
import json
from detectron2.structures import BoxMode
import itertools

from detectron2.engine import DefaultTrainer
from detectron2.config import get_cfg

cfg = get_cfg()

cfg.merge_from_file("/home/sangjoon/detectron2/configs/COCO-Detection/faster_rcnn_swint_T_FPN_3x_.yaml") cfg.DATALOADER.NUM_WORKERS = 4
cfg.MODEL.WEIGHTS = "/home/sangjoon/SwinT_detection2/real_white_weights/model_0015499.pth" # initialize from model zoo

trainer = DefaultTrainer(cfg)
trainer.resume_or_load(resume=False)

cfg.MODEL.ROI_HEADS.SCORE_THRESH_TEST = 0.5 # set the testing threshold for this model
predictor = DefaultPredictor(cfg)

KeyError Traceback (most recent call last)
in
1 cfg = get_cfg()
2 from swint import add_swint_config
----> 3 cfg.merge_from_file("/home/sangjoon/detectron2/configs/COCO-Detection/faster_rcnn_swint_T_FPN_3x_.yaml")
4 # cfg.DATALOADER.NUM_WORKERS = 4
5 cfg.MODEL.WEIGHTS = "/home/sangjoon/SwinT_detection2/real_white_weights/model_0015499.pth" # initialize from model zoo

~/detectron2/detectron2/config/config.py in merge_from_file(self, cfg_filename, allow_unsafe)
52
53 if loaded_ver == self.VERSION:
---> 54 self.merge_from_other_cfg(loaded_cfg)
55 else:
56 # compat.py needs to import CfgNode

~/.conda/envs/mmdetection/lib/python3.7/site-packages/fvcore/common/config.py in merge_from_other_cfg(self, cfg_other)
121 BASE_KEY not in cfg_other
122 ), "The reserved key '{}' can only be used in files!".format(BASE_KEY)
--> 123 return super().merge_from_other_cfg(cfg_other)
124
125 def merge_from_list(self, cfg_list: List[str]) -> Callable[[], None]:

~/.conda/envs/mmdetection/lib/python3.7/site-packages/yacs/config.py in merge_from_other_cfg(self, cfg_other)
215 def merge_from_other_cfg(self, cfg_other):
216 """Merge cfg_other into this CfgNode."""
--> 217 _merge_a_into_b(cfg_other, self, self, [])
218
219 def merge_from_list(self, cfg_list):

~/.conda/envs/mmdetection/lib/python3.7/site-packages/yacs/config.py in _merge_a_into_b(a, b, root, key_list)
476 if isinstance(v, CfgNode):
477 try:
--> 478 _merge_a_into_b(v, b[k], root, key_list + [k])
479 except BaseException:
480 raise

~/.conda/envs/mmdetection/lib/python3.7/site-packages/yacs/config.py in _merge_a_into_b(a, b, root, key_list)
489 root.raise_key_rename_error(full_key)
490 else:
--> 491 raise KeyError("Non-existent config key: {}".format(full_key))
492
493

KeyError: 'Non-existent config key: MODEL.SWINT'

How do I fix it??

I run training using detectron2

but this error comes up - KeyError: 'Non-existent config key: MODEL.SWINT'
How do I do???

Thank you!

xiaohu，非常感谢你伟大的工作，
我用在上面训练faster-Rcnn，训练几万步，ap为nan，我也采用你faster_rcnn_swint_T.pth作为pre-tained； 我也一直用detectron2，还是比较熟悉这个开源框架。找不到原因，请明示！！

感谢！

你好，我用这套代码训练Retinanet，FPN的参数设置为["stage2","stage3", "stage4", "stage5"]时，会提示错误
配置如下：

RuntimeError: Expected to have finished reduction in the prior iteration before starting a new one. 
This error indicates that your module has parameters that were not used in producing loss. 
You can enable unused parameter detection by (1) passing the keyword argument find_unused_parameters=True to torch.nn.parallel.DistributedDataParallel;
 (2) making sure all forward function outputs participate in calculating loss. I
f you already have done the above two steps, then the distributed data parallel module wasn’t able to locate the output tensors in the return value of your module’s forward function.

看提示是有模型有冗余，在训练过程中部分模型没有提供loss，参考建议（1）,添加find_unused_parameters=True后，代码可以正常运行。但我没有定位到是哪部分导致的，能麻烦帮忙解决一下是模型的哪部分造成的么？