o
oiq @ s d dl mZmZmZmZmZ d dlmZmZ d dl m
Z
d dl m Z m
Z
mZ d dlmZmZmZ ddlmZ ddlmZ G d d
d
eZG d d d eZG d
d deZdS ) )AnyCallableDictOptionalTuple) Optimizer lr_scheduler)
DataLoader)ModuleTensorstack)accuracymean_average_precisionmean_iou )Trainer)
Configurationc @ * e Zd ZdZdedeeef fddZdS )ImageClassifierTrainera Module to be used for image classification purposes.
The module subclasses :py:class:`~kornia.x.Trainer` and overrides the
:py:func:`~kornia.x.Trainer.evaluate` function implementing a standard
:py:func:`~kornia.metrics.accuracy` topk@[1, 5].
.. seealso::
Learn how to use this class in the following
`example `__.
argsreturnc G s< t |dkrt|\}}t||dd\}}| | dS )N )r )topk)top1top5)lenAssertionErrorr
item)selfr outtargetacc1acc5 r$ E/home/ubuntu/.local/lib/python3.10/site-packages/kornia/x/trainers.pycompute_metrics* s
z&ImageClassifierTrainer.compute_metricsN __name__
__module__ __qualname____doc__r r strfloatr& r$ r$ r$ r% r s r c @ r )SemanticSegmentationTrainera Module to be used for semantic segmentation purposes.
The module subclasses :py:class:`~kornia.x.Trainer` and overrides the
:py:func:`~kornia.x.Trainer.evaluate` function implementing IoU :py:func:`~kornia.metrics.mean_iou`.
.. seealso::
Learn how to use this class in the following
`example `__.
r r c G s@ t |dkrt|\}}t|d||jd }d| iS )Nr r iou)r r r argmaxshapemeanr )r r r r! r/ r$ r$ r% r& = s
z+SemanticSegmentationTrainer.compute_metricsNr' r$ r$ r$ r% r. 2 s
r. c s e Zd ZdZ ddedee dee dee dede j
d e d
e d ee
eed f d
ee ddf fdd
Zdede
eef defddZdedefddZdee
eef de
eef f fdd Z ZS )ObjectDetectionTrainera Module to be used for object detection purposes.
The module subclasses :py:class:`~kornia.x.Trainer` and overrides the
:py:func:`~kornia.x.Trainer.evaluate` function implementing IoU :py:func:`~kornia.metrics.mean_iou`.
.. seealso::
Learn how to use this class in the following
`example `__.
Nmodeltrain_dataloadervalid_dataloader criterion optimizer schedulerconfig num_classes callbacks).Nloss_computed_by_modelr c
s6 | d u ri } t |||||||| |
| _|| _d S )N)super__init__r= r; ) r r4 r5 r6 r7 r8 r9 r: r; r< r= __class__r$ r% r? P s
z ObjectDetectionTrainer.__init__samplec C s* | j r|jr||d |d S ||d S )Ninputr! )r= training)r r4 rB r$ r$ r% on_modeld s z ObjectDetectionTrainer.on_modelr c G sb | j r#t|d trtdd |d D S tt|d S | jd u r,t d| j| S )Nr c S s g | ]\}}|qS r$ r$ ).0_vr$ r$ r%
m z7ObjectDetectionTrainer.compute_loss..zD`criterion` should not be None if `loss_computed_by_model` is False.)
r=
isinstancedictr itemsr2 listsumr7 RuntimeError)r r r$ r$ r% compute_lossi s
z#ObjectDetectionTrainer.compute_lossc s t |d trcd|d v rcd|d v rcd|d v rct |d trcd|d v rcd|d v rctdd |d D dd |d D d d |d D d
d |d D d d |d D | jd d
\}}d| iS t j| S )Nr boxeslabelsscoresr c S g | ]}|d qS rR r$ rF ar$ r$ r% rI ~ rJ z:ObjectDetectionTrainer.compute_metrics..c S rU rS r$ rW r$ r$ r% rI rJ c S rU )rT r$ rW r$ r$ r% rI rJ c S rU rV r$ rW r$ r$ r% rI rJ c S rU rY r$ rW r$ r$ r% rI rJ gư>) n_classes thresholdmAP)rK rL r r; r r> r& )r r r\ rG r@ r$ r% r& s s(
z&ObjectDetectionTrainer.compute_metrics)NN)r( r) r* r+ r
r r r r r _LRSchedulerr intr r, r boolr? r rE rQ r r- r&
__classcell__r$ r$ r@ r% r3 E s<
2
r3 N)typingr r r r r torch.optimr r torch.utils.datar kornia.corer
r r kornia.metricsr
r r trainerr utilsr r r. r3 r$ r$ r$ r% s