Pre-trained TensorFlow models for object recognition

I often use this blog, also, as a personal diary because in some cases it is useful for me to record some sources to remind myself of interesting things that could be useful during work or experiments. On this occasion I decided to keep track of the pre-trained models that I have tested or that I would like to test, provided by third parties. I hope, therefore, that this article may also be useful to others. In this post, which I will update over time, I have started to collect a list of models useful for object recognition and tracking. In this first part I will list those trained on the datasets COCO , Kitti, Open Images,AVA v2.1 and iNaturalist Species. These models may be useful in general for those interested in these types of data categories, but they are certainly useful for starting to train new models. The list of models is organized according to the following scheme :

• the model name, which corresponds to the config file used for training,
• the link to download which contains an archive with the pre-trained model,
• The model speed expressed in ms and measured on 600X600 images
• mAp (mean Average Precision i.e. the mean precision or the “mAP score”; The established way to evaluate the performance of object detection models; this value is used in the PASCAL VOC, ImageNet and COCO challenges.
• Type of output supported (Boxes or Masks )

Each archive will contain:

• the graph proto (graph.pbtxt)
• the checkpoint files (model.ckpt.data-00000-of-00001, model.ckpt.index, model.ckpt.meta)
• the frozen graph proto (frozen_inference_graph.pb) which can be used directly for inference.
• the config file (pipeline.config) used to generate the graph model
• the exported TfLite model for mobile (model.tflite).

COCO-trained models

Model name	Speed (ms)	COCO mAP	Outputs
ssd_mobilenet_v1_coco	30	21	Boxes
ssd_mobilenet_v1_0.75_depth_coco	26	18	Boxes
ssd_mobilenet_v1_quantized_coco	29	18	Boxes
ssd_mobilenet_v1_0.75_depth_quantized_coco	29	16	Boxes
ssd_mobilenet_v1_ppn_coco	26	20	Boxes
ssd_mobilenet_v1_fpn_coco	56	32	Boxes
ssd_resnet_50_fpn_coco	76	35	Boxes
ssd_mobilenet_v2_coco	31	22	Boxes
ssd_mobilenet_v2_quantized_coco	29	22	Boxes
ssdlite_mobilenet_v2_coco	27	22	Boxes
ssd_inception_v2_coco	42	24	Boxes
faster_rcnn_inception_v2_coco	58	28	Boxes
faster_rcnn_resnet50_coco	89	30	Boxes
faster_rcnn_resnet50_lowproposals_coco	64		Boxes
rfcn_resnet101_coco	92	30	Boxes
faster_rcnn_resnet101_coco	106	32	Boxes
faster_rcnn_resnet101_lowproposals_coco	82		Boxes
faster_rcnn_inception_resnet_v2_atrous_coco	620	37	Boxes
faster_rcnn_inception_resnet_v2_atrous_lowproposals_coco	241		Boxes
faster_rcnn_nas	1833	43	Boxes
faster_rcnn_nas_lowproposals_coco	540		Boxes
mask_rcnn_inception_resnet_v2_atrous_coco	771	36	Masks
mask_rcnn_inception_v2_coco	79	25	Masks
mask_rcnn_resnet101_atrous_coco	470	33	Masks
mask_rcnn_resnet50_atrous_coco	343	29	Masks

Mobile models

Model name	Pixel 1 Latency (ms)	COCO mAP	Outputs
ssd_mobilenet_v3_large_coco	119	22.3	Boxes
ssd_mobilenet_v3_small_coco	43	15.6	Boxes

Kitti-trained models

Model name	Speed (ms)	Pascal mAP@0.5	Outputs
faster_rcnn_resnet101_kitti	79	87	Boxes

Open Images-trained models

Model name	Speed (ms)	Open Images mAP@0.5	Outputs
faster_rcnn_inception_resnet_v2_atrous_oidv2	727	37	Boxes
faster_rcnn_inception_resnet_v2_atrous_lowproposals_oidv2	347		Boxes
facessd_mobilenet_v2_quantized_open_image_v4 [^3]	20	73 (faces)	Boxes

Model name	Speed (ms)	Open Images mAP@0.5	Outputs
faster_rcnn_inception_resnet_v2_atrous_oidv4	425	54	Boxes
ssd_mobilenetv2_oidv4	89	36	Boxes
ssd_resnet_101_fpn_oidv4	237	38	Boxes

iNaturalist Species-trained models

Model name	Speed (ms)	Pascal mAP@0.5	Outputs
faster_rcnn_resnet101_fgvc	395	58	Boxes
faster_rcnn_resnet50_fgvc	366	55	Boxes

AVA v2.1 trained models

Model name	Speed (ms)	Pascal mAP@0.5	Outputs
faster_rcnn_resnet101_ava_v2.1	93	11	Boxes