small target detection no detections

Prerequisites

Please answer the following question for yourself before submitting an issue.

• I checked to make sure that this feature has not been requested already.

1. The entire URL of the file you are using

https://github.com/tensorflow/models/blob/master/research/object_detection/configs/tf2/faster_rcnn_resnet50_v1_640x640_coco17_tpu-8.config

2. Describe the feature you request

I want to use the the config above and my small target detection dataset to gain weight,after first train ,i think maybe the scale , hiight stride and width stride in grid_anchor_generator are so big that the network can't get valid feature,and there are no detections in the inference

 grid_anchor_generator {
    scales: [0.25, 0.5, 1.0, 2.0]
    aspect_ratios: [0.5, 1.0, 2.0]
    height_stride: 16
    width_stride: 16

after analyzing the dataset, i get the smallest scale,so I change to scales to :

  grid_anchor_generator {
    scales: [0.0078125,0.015625,0.03125,0.0625,0.125, 0.25, 0.5, 1.0]
    aspect_ratios: [0.5, 1.0, 2.0]
    height_stride: 2
    width_stride: 2

but the the loss of the network is like this

i can't find the astringent point.and tno detection as above.
anyone has done like this,does the od api exist a model to suit the small target detection ,please give me some advice ,tks.

3. Additional context

one of my dataset like this:

4. Are you willing to contribute it? (Yes or No)