image-model / .hydra /config.yaml
anthonymeo's picture
Upload folder using huggingface_hub
c6b353a verified
training_hyperparams:
resume: null
run_id: null
resume_path: null
resume_from_remote_sg_logger: false
ckpt_name: ckpt_latest.pth
lr_mode: CosineLRScheduler
lr_schedule_function: null
lr_warmup_epochs: 3
lr_warmup_steps: 1000
lr_cooldown_epochs: 0
warmup_initial_lr: 1.0e-06
step_lr_update_freq: null
cosine_final_lr_ratio: 0.1
warmup_mode: LinearEpochLRWarmup
lr_updates:
_target_: super_gradients.training.utils.utils.empty_list
pre_prediction_callback: null
optimizer: AdamW
optimizer_params:
weight_decay: 0.0001
load_opt_params: true
zero_weight_decay_on_bias_and_bn: true
loss: PPYoloELoss
criterion_params:
use_static_assigner: false
num_classes: ${num_classes}
reg_max: 16
ema: true
ema_params:
decay: 0.9
decay_type: threshold
beta: 15
train_metrics_list: []
valid_metrics_list:
- DetectionMetrics_050:
score_thres: 0.1
top_k_predictions: 300
num_cls: ${num_classes}
normalize_targets: true
post_prediction_callback:
_target_: super_gradients.training.models.detection_models.pp_yolo_e.PPYoloEPostPredictionCallback
score_threshold: 0.01
nms_top_k: 1000
max_predictions: 300
nms_threshold: 0.7
metric_to_watch: [email protected]
greater_metric_to_watch_is_better: true
launch_tensorboard: false
tensorboard_port: null
tb_files_user_prompt: false
save_tensorboard_to_s3: false
precise_bn: false
precise_bn_batch_size: null
sync_bn: true
silent_mode: false
mixed_precision: true
save_ckpt_epoch_list:
- 100
- 200
- 250
average_best_models: true
dataset_statistics: false
batch_accumulate: 1
run_validation_freq: 1
run_test_freq: 1
save_model: true
seed: 42
phase_callbacks: []
log_installed_packages: true
clip_grad_norm: null
ckpt_best_name: ckpt_best.pth
max_train_batches: null
max_valid_batches: null
sg_logger: base_sg_logger
sg_logger_params:
tb_files_user_prompt: false
launch_tensorboard: false
tensorboard_port: null
save_checkpoints_remote: false
save_tensorboard_remote: false
save_logs_remote: false
monitor_system: true
torch_compile: false
torch_compile_loss: false
torch_compile_options:
mode: reduce-overhead
fullgraph: false
dynamic: false
backend: inductor
options: null
disable: false
finetune: false
_convert_: all
max_epochs: 100
initial_lr: 0.0004
dataset_params:
data_dir: /workspace/final-final-3
dataset_name: ${dataset_name}
train_dataset_params:
data_dir: ${..data_dir}
dataset_name: ${..dataset_name}
split: train
input_dim:
- 640
- 640
cache_annotations: true
ignore_empty_annotations: false
transforms:
- DetectionMosaic:
input_dim: ${dataset_params.train_dataset_params.input_dim}
prob: 1.0
- DetectionRandomAffine:
degrees: 0.0
translate: 0.1
scales:
- 0.5
- 1.5
shear: 0.0
target_size: ${dataset_params.train_dataset_params.input_dim}
filter_box_candidates: false
wh_thr: 2
area_thr: 0.1
ar_thr: 20
border_value: 128
- DetectionHSV:
prob: 1.0
hgain: 5
sgain: 30
vgain: 30
- DetectionHorizontalFlip:
prob: 0.5
- DetectionPaddedRescale:
input_dim: ${dataset_params.train_dataset_params.input_dim}
- DetectionStandardize:
max_value: 255.0
- DetectionTargetsFormatTransform:
input_dim: ${dataset_params.train_dataset_params.input_dim}
output_format: LABEL_CXCYWH
class_inclusion_list: null
max_num_samples: null
with_crowd: false
verbose: 0
train_dataloader_params:
shuffle: true
batch_size: 12
min_samples: 512
num_workers: 4
drop_last: false
pin_memory: true
worker_init_fn:
_target_: super_gradients.training.utils.utils.load_func
dotpath: super_gradients.training.datasets.datasets_utils.worker_init_reset_seed
collate_fn: DetectionCollateFN
val_dataset_params:
data_dir: ${..data_dir}
dataset_name: ${..dataset_name}
split: valid
input_dim:
- 640
- 640
cache_annotations: true
ignore_empty_annotations: false
transforms:
- DetectionPaddedRescale:
input_dim: ${dataset_params.val_dataset_params.input_dim}
pad_value: 114
- DetectionStandardize:
max_value: 255.0
- DetectionTargetsFormatTransform:
input_dim: ${dataset_params.val_dataset_params.input_dim}
output_format: LABEL_CXCYWH
class_inclusion_list: null
max_num_samples: null
with_crowd: true
verbose: 0
val_dataloader_params:
batch_size: 16
num_workers: 4
drop_last: false
shuffle: false
pin_memory: true
collate_fn: CrowdDetectionCollateFN
_convert_: all
checkpoint_params:
load_checkpoint: false
load_backbone: false
checkpoint_path: null
external_checkpoint_path: null
source_ckpt_folder_name: null
strict_load:
_target_: super_gradients.training.sg_trainer.StrictLoad
value: no_key_matching
pretrained_weights: coco
checkpoint_num_classes: null
arch_params:
in_channels: 3
backbone:
NStageBackbone:
stem:
YoloNASStem:
out_channels: 48
stages:
- YoloNASStage:
out_channels: 96
num_blocks: 2
activation_type: relu
hidden_channels: 64
concat_intermediates: true
- YoloNASStage:
out_channels: 192
num_blocks: 3
activation_type: relu
hidden_channels: 128
concat_intermediates: true
- YoloNASStage:
out_channels: 384
num_blocks: 5
activation_type: relu
hidden_channels: 256
concat_intermediates: true
- YoloNASStage:
out_channels: 768
num_blocks: 2
activation_type: relu
hidden_channels: 384
concat_intermediates: false
context_module:
SPP:
output_channels: 768
activation_type: relu
k:
- 5
- 9
- 13
out_layers:
- stage1
- stage2
- stage3
- context_module
neck:
YoloNASPANNeckWithC2:
neck1:
YoloNASUpStage:
out_channels: 192
num_blocks: 2
hidden_channels: 192
width_mult: 1
depth_mult: 1
activation_type: relu
reduce_channels: true
neck2:
YoloNASUpStage:
out_channels: 96
num_blocks: 3
hidden_channels: 64
width_mult: 1
depth_mult: 1
activation_type: relu
reduce_channels: true
neck3:
YoloNASDownStage:
out_channels: 192
num_blocks: 2
hidden_channels: 192
activation_type: relu
width_mult: 1
depth_mult: 1
neck4:
YoloNASDownStage:
out_channels: 384
num_blocks: 3
hidden_channels: 256
activation_type: relu
width_mult: 1
depth_mult: 1
heads:
NDFLHeads:
num_classes: 80
reg_max: 16
heads_list:
- YoloNASDFLHead:
inter_channels: 128
width_mult: 0.75
first_conv_group_size: 0
stride: 8
- YoloNASDFLHead:
inter_channels: 256
width_mult: 0.75
first_conv_group_size: 0
stride: 16
- YoloNASDFLHead:
inter_channels: 512
width_mult: 0.75
first_conv_group_size: 0
stride: 32
bn_eps: 0.001
bn_momentum: 0.03
inplace_act: true
_convert_: all
num_classes: ${num_classes}
train_dataloader: roboflow_train_yolox
val_dataloader: roboflow_val_yolox
dataset_name: final-final-c2j0n-mdjfm/3
num_classes: 7
architecture: yolo_nas_m
load_checkpoint: false
result_path: null
resume: null
multi_gpu: false
num_gpus: 1
experiment_suffix: ''
experiment_name: ${architecture}_roboflow_${dataset_name}${experiment_suffix}
lr: 0.0004
batch_size: 12
val_batch_size: 16
ema: true
epochs: 100
num_workers: 4
ckpt_root_dir: null