Object detection using fastai Hello,
I am looking to create a fast ai object detection model on the deep fashion dataset but do not know how make it predict. How can I make my model output prediction on images. Here is the code:
The data is loaded using the COCO format as a json. Then I create the model and run the training like this:
def get_train_imgs(noop): return imgs
datablock = DataBlock(blocks=(ImageBlock, BBoxBlock, BBoxLblBlock),
splitter=RandomSplitter(),
get_items=get_train_imgs,
getters=getters,
item_tfms=item_tfms,
batch_tfms=batch_tfms,
n_inp=1)
model = resnet34()
encoder = create_body(model, pretrained=False)
get_c(dls)
%cd "Practical-Deep-Learning-for-Coders-2.0/Computer Vision"
from imports import *
arch = RetinaNet(encoder, get_c(dls), final_bias=-4)
create_head(124, 4)
arch.smoothers
arch.classifier
arch.box_regressor
ratios = [1/2,1,2]
scales = [1,2**(-1/3), 2**(-2/3)]
crit = RetinaNetFocalLoss(scales=scales, ratios=ratios)
def _retinanet_split(m): return L(m.encoder,nn.Sequential(m.c5top6, m.p6top7, m.merges, m.smoothers, m.classifier, m.box_regressor)).map(params)
learn = Learner(dls, arch, loss_func=crit, splitter=_retinanet_split)
learn.freeze()
learn.fit_one_cycle(2, slice(1e-5, 1e-4))
from fastai.vision.core import PILImage
image = PILImage.create('./000032.jpg')
prediction, _, _ = learn.predict(image)
predict is outputing the following error: TypeError: clip_remove_empty() missing 2 required positional arguments: ‘bbox’ and ‘label’
I understand that I am not suppose to use .predict but I cant find any reference on what else to use.