9i=SSKrSSKJrJr SSKrSSKJrJr SSKJ r SSK J r SSK J r SSKJr SS KJr SS KJr S \lS\lS\l"S S \5rg)N)AnyDict)Image ImageFile)create_transform) transforms) load_image)ModeKeys)OfaBasePreprocessor) RandomAugmentTc^\rSrSrSr\R 4U4SjjrS\\ \ 4S\\ \ 44Sjr S\\ \ 4S\\ \ 44Sjr S\\ \ 4S\\ \ 44Sjr S rU=r$) "OfaImageClassificationPreprocessorz1 OFA preprocessor for image classification task. c.>[[U] "XU/UQ70UD6 UR[R :wa[ R"S[ R"URUR4[ RRS9[ R"5[ R"URURS9/5Ulg[#URSSSSSS S URURS 9 Ul[ R"[$R&"S S /UR R SSUR R S/[)SSS/SQS9/UR R SS/55Ulg)zpreprocess the data Args: cfg(modelscope.utils.config.ConfigDict) : model config model_dir (str): model path, mode: preprocessor mode (model mode) c$URS5$NRGBconvertimages q/var/www/html/land-doc-ocr/venv/lib/python3.13/site-packages/modelscope/preprocessors/ofa/image_classification.py=OfaImageClassificationPreprocessor.__init__..+s emmE2) interpolation)meanstdTg?zrand-m9-mstd0.5-inc1bicubicg?pixelr ) input_size is_training color_jitter auto_augmentrre_probre_modere_countrrc X-$N)xys rrr?saerc$URS5$rrrs rrrAs emmE&:rN) Identity AutoContrastEqualize Brightness SharpnessShearXShearY TranslateX TranslateYRotate)isPILaugs)superr__init__moder TRAINrComposeResizepatch_image_sizeInterpolationModeBICUBICToTensor Normalizerrpatch_resize_transformr functoolsreducer )selfcfg model_dirr@argskwargs __class__s rr?+OfaImageClassificationPreprocessor.__init__s 0   T D48 D">"F"FH##%$$$))B >+D '+;00  3'YYHH +D '+5*<*<  !3://::2A>00;;A>?%"&"  //::12>#6+D 'rdatareturncUR[R:XaURU5$UR U5$r*)r@r rA_build_train_sample_build_infer_sample)rLrSs r__call__+OfaImageClassificationPreprocessor.__call__Ss4 99 &++D1 1++D1 1rc>URU5nSRUS5nUSS0US'URUSS9US'[R"UR USSS /5US 'UR b[R"[US 5[UR545R5n[[US 55H>nUS SUS -R5nUR RU5nS XEU'M@ XBS 'U$)a Building training samples. step 1. Preprocess the data using the logic of `_build_infer_sample` and make sure the label data in the result. step 2. Preprocess the label data. Contains: - add ` ` before the label value and add `ref_dict` value - tokenize the label as `target` value without `bos` token. - add `bos` token and remove `eos` token of `target` as `prev_output_tokens`. - add constraints mask. Args: data (`Dict[str, Any]`): Input data, should contains the key of `image`, `prompt` and `label`, `image` refers the image input data, `prompt` refers the text input data the `label` is the supervised data for training. Return: A dict object, contains source, image, mask, label, target tokens, and previous output tokens data. z {}labelg?ref_dictF)add_bostargetNprev_output_tokensr Tconstraint_mask)rWformat tokenize_texttorchcatbos_itemconstraint_triezeroslentgt_dictboolrangetolistget_next_layer)rLrSsampler^raiconstraint_prefix_tokenconstraint_nodess rrV6OfaImageClassificationPreprocessor._build_train_sampleYs;())$/fWo.$Wos3z--fe-Dx',yy ]]F8,Sb1 2(4#$    +#kk3v6J/K+L+.t}}+=+?@@D 3v&:;<=*0(+**01q5+228&((#'#7#7#F#F+$- 7;"#34 > )8$ % rcURXRS5nURU5nURRR SS5nUR U5nUU[R"S/5URS.nSUR;a'URSU;aXRSUS'U$)a Building inference samples. step 1. Get the pillow image. step 2. Do some transforms for the pillow image as the image input, such as resize, normalize, to tensor etc. step 3. Tokenize the prompt as text input. step 4. Determine Whether or not to add labels to the sample. Args: data (`Dict[str, Any]`): Input data, should contains the key of `image` and `prompt`, the former refers the image input data, and the later refers the text input data. Return: A dict object, contains source, image, mask and label data. rpromptz what does the image describe?T)source patch_image patch_maskdecoder_prompttextr[) get_img_pil column_maprIrMmodelgetrcrdtensorrf)rLrSrrwruinputsros rrW6OfaImageClassificationPreprocessor._build_infer_samples   oog&>!?@11%8 ##H.NO##F+&,,v."mm   T__ $)@D)H"??6#:;F7O r)rI)__name__ __module__ __qualname____firstlineno____doc__r INFERENCEr?rstrrrXrVrW__static_attributes__ __classcell__)rQs@rrrs((8t2T#s(^2S#X2 &S#X&4S>&PS#X4S>rr)rJtypingrrrdPILrr timm.datar torchvisionrmodelscope.preprocessors.imager modelscope.utils.constantr baser utils.vision_helperr LOAD_TRUNCATED_IMAGESMAX_IMAGE_PIXELSrr+rrrsJ &"5.%."& ! I)<Ir