CNN with Embedding Transformers for Person Reidentification

Table 2

More detail of TIC (based on Market-1501 and input resize to 384 × 192).

Layer name	Output size	50-layer
conv1		, 64, stride 2
conv2_x		, max pool, stride 2
		RT		Block x1
conv3_x
RT		Block x1
conv4_x
conv5_x
		PGAP, 751-d fc, softmax