Dataset mismatch while training

Hi, I have object detection data labeled using Roboflow, and I’m working with TensorFlow. I downloaded the dataset in TFRecord format, and the labels are also provided as a CSV file. However, each image is repeated multiple times—once for each bounding box annotation, as shown in the example. To address this, I wrote a script that consolidates all labels for a single image into one entry. But when I try to train the model using this modified dataset, I encounter a mismatch error. How can I resolve this issue?
csv file

i have 1605 images
each image hass different numbers of bboxes and classes
for example in one of the batches i get
16 images
(16, 23, 4) bbox
(16, 23, 1) class
noteice that each batch is padded to get uniform shape
when i fit my dataset i get this error

how to solve this problem

Does the algorithm you plan to use support detecting more than one object per image? If so, how? Your ground truth label data will need to mimic the shape of your network output, no?