^jddlmZddlmZddlmZddlmZddl Z ddl m Z ddlmZddlmZmZmZmZddlmZddlmZdd lmZdd lmZeZGd d Zy) ) annotations)Path)AnyN)Axes) BoxAnnotatorColor DetectionsLabelAnnotator)Tensor) DataLoader)box_cxcywh_to_xyxy) get_loggercdeZdZdZ d ddZddZe d dZy) DatasetGridSaveraaUtility for saving 3x3 image grids to visualize augmentation effects. Args: data_loader: Dataloader of the dataset to sample images from. output_dir: Directory where grid images will be saved. max_batches: Number of batches to draw samples from. dataset_type: Dataset split label, e.g. ``'train'`` or ``'val'``. cv||_||_||_||_|jj ddy)NT)parentsexist_ok) data_loader output_dir max_batches dataset_typemkdir)selfrrrrs e/var/www/ramen.bs-engineer-server.com/venv/lib/python3.12/site-packages/rfdetr/datasets/save_grids.py__init__zDatasetGridSaver.__init__#s:'$&( dT:c tjgdgd}td}ttj dd}t |jD]/\}\}}||jk\rntjddd \}}|j|jd ||j}d } t t|j|D](\} \} } | d k\rn|j!| | || |||*t#| d D]} || j%d|j'tj(|j*|jd|dz dtj,2t.j1d|jd|j*j3y)zCreate and save image grids to ``output_dir``. Each grid is a 3x3 JPEG containing up to 9 images from a single batch, with bounding boxes and class labels drawn on top. )g:ܟwg$I$IggE#)g!:ܟw@gm۶m@grq@)meanstd) thicknessg?) text_color text_scale text_padding) r&)figsizez dataset, batch r off_batchz _grid.jpg)dpizSaved z! grids with augmented images to: N)T Normalizerr rBLACK enumeraterrpltsubplotssuptitlerflattenziptensors_annotate_and_plotrangeaxis tight_layoutsavefigrcloseloggerinforesolve) r inv_normalize box_annotatorlabel_annotator batch_idxsampletargetfigaxes sample_index single_image single_targetis r save_gridzDatasetGridSaver.save_grid,s  A1 %q1 ({{  ,5T5E5E+F  'I'D,,, Q8ykJ K<<>DL?HV^^]cId?e ; ;|]1$'' -l1C]Tacr <+ $Q U# $     KKT->->,?vi[PY*ZZ`c d IIK+ .  fT..//PQUQ`Q`QhQhQjPklmrc ddlm}|d}t|tr|j j }t |dt |d} }||} t| tr,| j j j} |jtj| jddddddzjtj} t|d dkDrN|d } t| tr@| j j jjt } ntj| t } |d }t|tr|j j }n|}tj|Dcgc]+}t!|}|d| z|d|z|d| z|d |zg-c}}tj" }t%|| }| Dcgc] }t'|}}|j)| |} |j)| ||} |j+| |j-dycc}}wcc}w)a|De-normalize a single image tensor, annotate it with boxes and labels, and plot it on ``ax``. Args: single_image: Normalized image tensor of shape ``(C, H, W)``. single_target: Target dict with keys ``'size'``, ``'boxes'`` (cx,cy,wh normalized), and ``'labels'``. ax: Matplotlib axis to plot the annotated image on. inv_normalize: Inverse normalization transform to convert the tensor back to pixel values. box_annotator: ``BoxAnnotator`` instance for drawing bounding boxes. label_annotator: ``LabelAnnotator`` instance for drawing class labels. r)Imagesizer gg?boxeslabels)dtyper")xyxyclass_id)scene detections)rWrXrSr)N)PILrN isinstancer detachcpuintnumpy fromarraynpclip transposeastypeuint8lenasarrayr float32r strannotateimshowr9)rIrJaxr@rArBPILImage resized_sizehwde_normalized_imgrW labels_tensor class_idsrR boxes_iterboxbrUrXcrSs rr7z#DatasetGridSaver._annotate_and_plotVs%& *$V, lF +'..0446L<?#Sa%91),7 ' 0 1 8 8 : > > @ F F H ""BGG,=,G,G1a,PRUWZ$[^a$a#i#ijljrjr#st }W% & *)(3M-0)002668>>@GGL JJ}C@ !'*E%("\\^//1 " ::EOscZlmpZqTU!A$(AaD1HadQh!q9sjjD$ BJ&/0c!f0F0!**:*NE#,,5ZX^,_E % t1s =0I4 I:N)r"train) rr rrrr]rrhreturnNone)rxry)rIr rJzdict[str, Any]rkrr@z T.NormalizerArrBr rxry)__name__ __module__ __qualname____doc__rrL staticmethodr7rrrrsdk;%;37;FI;]`; ;(nT55%5 5# 5 $ 5 ( 5 55rr) __future__rpathlibrtypingrmatplotlib.pyplotpyplotr1r^r`torchvision.transforms transformsr-matplotlib.axesr supervisionrrr r torchr torch.utils.datar rfdetr.util.box_opsr rfdetr.util.loggerrr=rrrrrsA #" GG'2) ssr